CN106302614A - A kind of data distributing method and device - Google Patents
A kind of data distributing method and device Download PDFInfo
- Publication number
- CN106302614A CN106302614A CN201510317846.2A CN201510317846A CN106302614A CN 106302614 A CN106302614 A CN 106302614A CN 201510317846 A CN201510317846 A CN 201510317846A CN 106302614 A CN106302614 A CN 106302614A
- Authority
- CN
- China
- Prior art keywords
- llc
- node
- home
- cache lines
- enumerator
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/56—Provisioning of proxy services
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The embodiment of the invention discloses a kind of data distributing method and device, the method comprises the steps that the requesting node node to local Home last level cache LLC place sends request bag；When the node at described Home LLC place receives described request bag, identify whether the transport overhead between described requesting node and the node at described Home LLC place exceedes Preset Transfer overhead thresholds according to described request bag；When described transport overhead exceedes Preset Transfer overhead thresholds, the cache lines at the data place asked by described request bag copies in the LLC of described requesting node from described Home LLC.The embodiment of the present invention can reduce the transport overhead in many-core framework.
Description
Technical field
The present invention relates to communication technical field, particularly relate to a kind of data distributing method and device.
Background technology
Along with the expansion of chip-scale, increasing chip uses many-core framework.But, current many-core
The data distribution mechanism of the last level cache (Last Level Cache, LLC) used in framework is mainly data
It is only remained in this locality (Home) LLC that chip is local.When chip interior joint needs to use certain data, should
The node node to Home LLC place sends request of data, and the node at Home LLC place receives this number
After request, these data of storage in Home LLC are replicated data to this node.But, due to crowd
Core framework chips can exist a lot of node, the node at so different nodes and Home LLC place it
Between transmission path can difference, so will necessarily exist between the node at some nodes and Home LLC place
Transmission path can be distant, expense meeting during the node-node transmission data at so this node and Home LLC place
Bigger.Thus, can there is, in current many-core framework, the problem that transport overhead is excessive.
Summary of the invention
The invention provides a kind of data distributing method and device, it is possible to reduce the transport overhead in many-core framework.
First aspect, the present invention provides a kind of data distributing method, including:
The requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify according to described request bag
Whether the transport overhead between the node at described requesting node and described Home LLC place exceedes Preset Transfer
Overhead thresholds；
When described transport overhead exceedes Preset Transfer overhead thresholds, the data institute that described request bag is asked
Cache lines copy to from described Home LLC in the LLC of described requesting node.
In the first possible implementation of first aspect, described requesting node is to Home LLC place
Node send request bag during described request bag often in a node, described request bag on sheet
Network counter adds N, and wherein, described N is the integer more than or equal to 1；
Described according between the described request bag described requesting node of identification and the node at described Home LLC place
Transport overhead whether exceed Preset Transfer overhead thresholds, including:
Judge whether the currency of the described network-on-chip enumerator in described request bag is more than Preset Transfer expense
Threshold value.
In conjunction with the first possible implementation of first aspect, the realization that the second in first aspect is possible
In mode, described request bag is the read request packet that the node to described Home LLC place reads data；
Described method also includes:
When the node at described Home LLC place receives described request bag, default being used for is added up institute
The local enumerator stating the reading times that requesting node reads the first cache lines adds N, and wherein, described first delays
Deposit the cache lines at the data place that read request packet described in behavior is read；
Whether the currency of the described described network-on-chip enumerator judged in described request bag is more than Preset Transfer
Overhead thresholds, including:
Judge the currency of described network-on-chip enumerator in described request bag and working as of described local enumerator
Whether front value sum is more than Preset Transfer overhead thresholds；
Described when described transport overhead exceedes Preset Transfer overhead thresholds, that described request bag is asked number
Copy in the LLC of described requesting node from described Home LLC according to the cache lines at place, including:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator are more than described
During Preset Transfer overhead thresholds, described first cache lines is copied to described request from described Home LLC
In the LLC of node and level cache.
In conjunction with the implementation that the second of first aspect is possible, in the third possible realization of first aspect
In mode, described method also includes:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator less than or
During equal to described Preset Transfer overhead thresholds, described first cache lines is copied to from described Home LLC
In the level cache of described requesting node.
In conjunction with the first possible implementation of first aspect, in the 4th kind of possible realization of first aspect
In mode, described request bag is the write request bag that the second cache lines to described Home LLC carries out write operation；
Whether the currency of the described described network-on-chip enumerator judged in described request bag is more than Preset Transfer
Overhead thresholds, including:
Judge the currency of the described described network-on-chip enumerator asked in bag and the currency of far-end enumerator
Whether sum is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is for pre-setting for adding up
Distant-end node reads the reading times of described second cache lines, and described distant-end node is except described Home in chip
Node outside the node at LLC place；
Described when described transport overhead exceedes Preset Transfer overhead thresholds, that described request bag is asked number
Copy in the LLC of described requesting node from described Home LLC according to the cache lines at place, including:
When the currency of described network-on-chip enumerator and the currency sum of described far-end enumerator are more than presetting
During transport overhead threshold value, amended second cache lines is copied to described request from described Home LLC
In the LLC of node and level cache, wherein, write request described in described amended second caching behavior response
The cache lines that bag carries out write operation to described second cache lines and generates.
In conjunction with the 4th kind of possible implementation of first aspect, in the 5th kind of possible realization of first aspect
In mode, described method also includes:
By invalid for the copy of described second cache lines of storage in described chip, wherein, described second cache lines
Copy data of the second cache lines storage described in when being not respond described write request bag.
Second aspect, the present invention provides a kind of data distribution device, including: request unit, recognition unit and
First copied cells, wherein:
Described request unit, for controlling the requesting node node to local Home last level cache LLC place
Send request bag；
Described recognition unit, is used for when the node at described Home LLC place receives described request bag,
Identify that the transmission between described requesting node and the node at described Home LLC place is opened according to described request bag
Whether pin exceedes Preset Transfer overhead thresholds；
Described first copied cells, for when described transport overhead exceedes Preset Transfer overhead thresholds, by institute
The cache lines stating the data place that request bag is asked copies to described requesting node from described Home LLC
LLC in.
In the first possible implementation of second aspect, described requesting node is to Home LLC place
Node send request bag during described request bag often in a node, described request bag on sheet
Network counter adds N, and wherein, described N is the integer more than or equal to 1；
Whether described recognition unit for judging the currency of the described network-on-chip enumerator in described request bag
More than Preset Transfer overhead thresholds.
In conjunction with the first possible implementation of second aspect, the realization that the second in second aspect is possible
In mode, described request bag is the read request packet that the node to described Home LLC place reads data；
Described device also includes:
Counting unit, for when the node at described Home LLC place receives described request bag, by pre-
If add N for the local enumerator adding up the reading times that described requesting node reads the first cache lines, its
In, the cache lines at the data place that read request packet described in described first caching behavior is read；
Described recognition unit is for judging currency and the institute of the described network-on-chip enumerator in described request bag
Whether state the currency sum of local enumerator more than Preset Transfer overhead thresholds；
Described first copied cells is used for the currency when described network-on-chip enumerator and described local enumerator
Currency sum more than described Preset Transfer overhead thresholds time, by described first cache lines from described Home
LLC copies in LLC and the level cache of described requesting node.
In conjunction with the implementation that the second of second aspect is possible, in the third possible realization of second aspect
In mode, described device also includes:
Second copied cells, is used for the currency when described network-on-chip enumerator and described local enumerator
When currency sum is less than or equal to described Preset Transfer overhead thresholds, by described first cache lines from described
Home LLC copies in the level cache of described requesting node.
In conjunction with the implementation that the second of second aspect is possible, in the 4th kind of possible realization of second aspect
In mode, described request bag is the write request bag that the second cache lines to described Home LLC carries out write operation；
Described recognition unit is for judging that the currency of the described network-on-chip enumerator in described request bag is with remote
Whether the currency sum of end enumerator is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is
Pre-setting the reading times reading described second cache lines for adding up distant-end node, described distant-end node is
Node in addition to the node at described Home LLC place in chip；
Described first copied cells is used for when the currency of described network-on-chip enumerator and working as of far-end enumerator
When front value sum is more than Preset Transfer overhead thresholds, by amended second cache lines from described Home LLC
In copy in LLC and the level cache of described requesting node, wherein, described amended second cache lines
The cache lines generated for described second cache lines is carried out the described write request bag of response write operation.
In conjunction with the 4th kind of possible implementation of second aspect, in the 5th kind of possible realization of second aspect
In mode, described device also includes:
Invalid unit, the copy of described second cache lines for storing in described chip is invalid, wherein,
The data of the second cache lines storage described in when the copy of described second cache lines is not respond described write request bag.
In technique scheme, the requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify described according to described request bag
Whether the transport overhead between the node at requesting node and described Home LLC place exceedes Preset Transfer expense
Threshold value；When described transport overhead exceedes Preset Transfer overhead thresholds, the data that described request bag is asked
The cache lines at place copies in the LLC of described requesting node from described Home LLC.Due to can be by
The cache lines at the data place of requesting node request is cached in the LLC of this requesting node, so this request joint
Point, when using these data, directly obtains just, without removing Home LLC from the LLC of this node
The node at place obtains, such that it is able to the transport overhead reduced in many-core framework.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to enforcement
In example or description of the prior art, the required accompanying drawing used is briefly described, it should be apparent that, describe below
In accompanying drawing be only some embodiments of the present invention, for those of ordinary skill in the art, do not paying
On the premise of going out creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is the schematic flow sheet that the embodiment of the present invention provides a kind of data distributing method；
Fig. 2 is a kind of transmission path schematic diagram asking bag transmission that the embodiment of the present invention provides；
Fig. 3 is the schematic flow sheet of the another kind of data distributing method that the embodiment of the present invention provides；
Fig. 4 is the schematic diagram of data cached copy in a kind of many-core framework that the embodiment of the present invention provides；
Fig. 5 is the structural representation of a kind of data distribution device that the embodiment of the present invention provides；
Fig. 6 is the structural representation of the another kind of data distribution device that the embodiment of the present invention provides；
Fig. 7 is the structural representation of the another kind of data distribution device that the embodiment of the present invention provides；
Fig. 8 is the chip structure schematic diagram of a kind of many-core framework that the embodiment of the present invention provides.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clearly
Chu, be fully described by, it is clear that described embodiment be only a part of embodiment of the present invention rather than
Whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creation
The every other embodiment obtained under property work premise, broadly falls into the scope of protection of the invention.
Referring to Fig. 1, Fig. 1 is the schematic flow sheet that the embodiment of the present invention provides a kind of data distributing method, as
Shown in Fig. 1, comprise the following steps:
101, the requesting node node to Home LLC place sends request bag.
Wherein, the node except above-mentioned Home LLC place during above-mentioned requesting node can be many-core framework it
Outer arbitrary node, above-mentioned Home LLC can be for data cached LLC in this many-core framework, its
In, the data of this Home LLC caching can be read by arbitrary node in this many-core framework, it addition, this crowd
Arbitrary node in core framework can also carry out write operation to the data of caching in Home LLC.
In the present embodiment, above-mentioned request bag can be read request packet, i.e. to Home LLC read operation；Or
Above-mentioned request bag can also is that write request bag, i.e. write operation in Home LLC.
102, when the node at described Home LLC place receives described request bag, according to described request bag
Identify whether the transport overhead between the node at described requesting node and described Home LLC place exceedes default
Transport overhead threshold value.
Above-mentioned transport overhead can be understood as the transmission road between the node at requesting node and Home LLC place
Footpath distance, or can be understood as the transmission path institute outside the node at requesting node and Home LLC place
The nodes of process；Or above-mentioned transport overhead it is also understood that and sends above-mentioned request bag extremely into requesting node
The duration that the node at Home LLC place is consumed, by the transport overhead of reflection request bag a length of time this.
103, when described transport overhead exceedes Preset Transfer overhead thresholds, the number that described request bag is asked
Copy to from described Home LLC according to the cache lines at place in the LLC of described requesting node.
In step 103 when transport overhead exceedes Preset Transfer overhead thresholds, just explanation requesting node with
The node-node transmission expense at Home LLC place is relatively big, if requesting node reads data in Home LLC
Time, then the transport overhead of network-on-chip (Network on Chip, NoC) will be very big, and step 103
Copy data in the LLC of requesting node, so when requesting node uses these data, directly can be from
The LLC of requesting node reads, such that it is able to the transport overhead reducing NoC will be the biggest；Wherein, should
NoC may be used for the network of multi-node communication on the chip of many-core framework.
Asked to read it addition, the cache lines at data place that above-mentioned request bag is asked can be request bag
The cache lines at data place of operation, or above-mentioned request bag asked to carry out the delaying of data place of write operation
Deposit row, and above-mentioned cache lines can be one or more cache lines.
Such as: the node 202 at requesting node 201 and Home LLC place, requesting node as shown in Figure 2
Transmission transmission path 203 as shown in Figure 2, path between the node 202 at 201 and Home LLC places.
As shown in Figure 2, between the node 202 at requesting node 201 and Home LLC place transmit path away from
From farther out, the transmission i.e. transmitting data between the node 202 at requesting node 201 and Home LLC place is opened
Pin can be bigger, and passes through above-mentioned steps by the data place required for cache request node in Home LLC
Cache lines copies to the LLC of requesting node 201, and when so next time uses these data, requesting node is just not required to
Send request bag, such that it is able to reduce transport overhead.
It addition, the cache lines at the data place asked by request bag in step 103 is from described Home LLC
In copy to the LLC of requesting node can be this cache lines replicated according to the transmission path of transmission request bag
LLC to requesting node；Or the cache lines at the data place asked by request bag in step 103 is from institute
State and Home LLC copies in the LLC of requesting node can be that the data processor of chip is by this cache lines
It is copied directly in the LLC of requesting node.It addition, above-mentioned duplication can be the copy of duplicate cache row, on
Stating cache lines can be Cacheline, or cache line.
In the present embodiment, said method can apply in the chip of any many-core framework, or is applied to appoint
Meaning includes the smart machine of the chip of many-core framework.
In the present embodiment, the requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify described according to described request bag
Whether the transport overhead between the node at requesting node and described Home LLC place exceedes Preset Transfer expense
Threshold value；When described transport overhead exceedes Preset Transfer overhead thresholds, the data that described request bag is asked
The cache lines at place copies in the LLC of described requesting node from described Home LLC.Due to can be by
The cache lines at the data place of requesting node request is cached in the LLC of this requesting node, so this request joint
Point, when using these data, directly obtains just, without removing Home LLC from the LLC of this node
The node at place obtains, such that it is able to the transport overhead reduced in many-core framework.
Refer to the schematic flow sheet that Fig. 3, Fig. 3 are the another kind of data distributing methods that the embodiment of the present invention provides,
As it is shown on figure 3, comprise the following steps:
301, the requesting node node to Home LLC place sends request bag, wherein, described requesting node
Node to Home LLC place sends described request during request is wrapped and wraps often through a node,
In described request bag, network-on-chip enumerator adds N, and wherein, described N is the integer more than or equal to 1.
Wherein, the requesting node node to Home LLC place send request bag be by requesting node with
Node-node transmission between the node at Home LLC place, transmission path 202 as shown in Figure 2.So should
Request bag often through a node, this node just by request wrap in network-on-chip enumerator add N (such as:
Add 1), when this request bag that the node at Home LLC place obtains, it is possible to by this request bag
This request bag of currency identification of network-on-chip counting have passed through many each and every one node, the i.e. transmission of this request bag and opens
Pin.It addition, when requesting node sends above-mentioned request bag, can be the network-on-chip counting in this request bag
Device sets an initial value, such as: 0.It addition, above-mentioned network-on-chip enumerator can be defined as (NoC Count,
NC), this NC is for adding up the transmission range that request wraps in network-on-chip, and this NC may be located at request bag
Packet header in.
302, when the node at described Home LLC place receives described request bag, it is judged that described request bag
In the currency of described network-on-chip enumerator whether more than Preset Transfer overhead thresholds.
Wherein, above-mentioned Preset Transfer overhead thresholds can be that user is set in advance, such as: 5,10,20 etc.,
Specifically can set according to the quantity of many-core framework interior joint, or according to the transmission path of many-core framework and
Set.
303, when the currency of network-on-chip enumerator is more than Preset Transfer overhead thresholds, institute is wrapped in described request
The cache lines at the data place of request copies in the LLC of described requesting node from described Home LLC.
When the currency of network-on-chip enumerator is more than Preset Transfer overhead thresholds, just explanation requesting node with
In Home LLC, between the node at place, transmission range is farther out.
In this enforcement just, above-mentioned request bag can be that the node to described Home LLC place reads data
Read request packet, i.e. requesting node read certain data to Home LLC request.Such as: the one-level of requesting node
After caching (L1 Cache) sends disappearance (miss), and the LLC in requesting node does not read institute
During the data needed, requesting node just can send to the node at the Home LLC place being cached with these data
Read request packet.Wherein, requesting node can be by directory search to the data being cached with required for requesting node
Home LLC.
In this embodiment, said method can also comprise the steps:
When the node at described Home LLC place receives described request bag, default being used for is added up institute
The local enumerator stating the reading times that requesting node reads the first cache lines adds N, and wherein, described first delays
Deposit the cache lines at the data place that read request packet described in behavior is read；
Step 302 may include that
Judge the currency of described network-on-chip enumerator in described request bag and working as of described local enumerator
Whether front value sum is more than Preset Transfer overhead thresholds；
Step 303 may include that
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator are more than described
During Preset Transfer overhead thresholds, described first cache lines is copied to described request from described Home LLC
In the LLC of node and level cache.
Wherein, above-mentioned local enumerator is set in advance, and local enumerator is caching behavior granularity,
And can be that each cache lines sets multiple local enumerator, each local enumerator is for one joint of statistics
Point reads the number of times of this cache lines, such as: can set M local enumerator for above-mentioned first cache lines,
Wherein, a number of times reading the first cache lines for adding up above-mentioned requesting node deposited by M local enumerator.
It addition, above-mentioned local enumerator can be defined as (Home Count, HC), this HC is for adding up this locality
Cache lines is accessed for frequency.
This embodiment can realize add up requesting node and read the number of times of above-mentioned first cache lines, work as request
Node read the number of times of the first cache lines and request bag the number of nodes sum of process more than described Preset Transfer
During overhead thresholds, described first cache lines is copied to from described Home LLC the LLC of described requesting node
With in level cache.So can realize reading the first cache lines number of times with requesting node and request bag is passed through
Nodes the two dimension decide whether to copy to the first cache lines the LLC of requesting node.Because working as
Requesting node read the number of times of the first cache lines more time, illustrate that the first cache lines is heavier for requesting node
, such that it is able to the first cache lines copies to the LLC of requesting node, and requesting node is repeatedly from Home
LLC reads the first cache lines, thus causes bigger transport overhead；It addition, as requesting node and Home LLC
Between transmission path farther out time, requesting node reads the first cache lines from Home LLC and will also result in bigger
Transport overhead.
It addition, the first cache lines can be copied to by present embodiment in the level cache of requesting node, this
Sample requesting node can realize when level cache sends miss the data required for level cache are copied to one
In level caching, thus solve miss problem.
In present embodiment, said method can also comprise the steps:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator less than or
During equal to described Preset Transfer overhead thresholds, described first cache lines is copied to from described Home LLC
In the level cache of described requesting node.
This embodiment can realize requesting node and can realize level cache when level cache sends miss
Required data copy in level cache, thus solve miss problem.Further, since network-on-chip meter
The currency of number device is less than or equal to described Preset Transfer overhead thresholds with the currency sum of local enumerator
Time, illustrate that requesting node reads the number of times of the first cache lines less, or requesting node and Home LLC institute
Node between transmission path close together, it is possible to the first cache lines is not copied to request joint
In the LLC of point.
In the present embodiment, above-mentioned request bag is that the second cache lines to described Home LLC carries out write operation
Write request bag；
Whether the currency of the above-mentioned described network-on-chip enumerator judged in described request bag is more than Preset Transfer
The step of overhead thresholds, may include that
Judge the currency of the described described network-on-chip enumerator asked in bag and the currency of far-end enumerator
Whether sum is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is for pre-setting for adding up
Distant-end node reads the reading times of described second cache lines, and described distant-end node is except described Home in chip
Node outside the node at LLC place；
Step 303 may include that
When the currency of described network-on-chip enumerator and the currency sum of far-end enumerator are more than Preset Transfer
During overhead thresholds, amended second cache lines is copied to described requesting node from described Home LLC
LLC and level cache in, wherein, described amended second caching behavior response described in write request bag pair
The cache lines that described second cache lines carries out write operation and generates.
Said distal ends node is all nodes in chip in addition to the node at described Home LLC place, or
Person's said distal ends node be in chip in addition to the node and above-mentioned requesting node at described Home LLC place
All nodes.So can count the second cache lines by said distal ends enumerator to be read or accessed
Number of times or frequency.It addition, said distal ends enumerator can be defined as (Remote Count, RC).Example
As: when distant-end node reads the second cache lines, this RC adds 1.
It addition, in this embodiment, request bag can also be carried out by above-mentioned request bag according to consistency protocol
The corresponding information of data of write operation be sent to the node at Home LLC place.
It addition, said method can also comprise the steps: in this embodiment
Respond above-mentioned request bag, the second cache lines is carried out write operation.
This write operation can include amendment, add or the write operation such as deletion.By second after this write operation
Cache lines is above-mentioned amended second cache lines.
This embodiment can realize when network-on-chip enumerator currency and far-end enumerator currency it
With during more than Preset Transfer overhead thresholds, the cache lines of carried out for requesting node write operation is copied to request joint
In the LLC of point
In this embodiment, said method can also comprise the steps:
By invalid for the copy of described second cache lines of storage in described chip, wherein, described second cache lines
Copy data of the second cache lines storage described in when being not respond described write request bag.
Such as: when network-on-chip enumerator currency and far-end enumerator currency sum less than or etc.
When Preset Transfer overhead thresholds, amended second cache lines is not copied to from described Home LLC
In the LLC of described requesting node and level cache, simply by described second cache lines of storage in described chip
Copy invalid.Because in the chips in addition to above-mentioned Home LLC is buffered in the copy of the second cache lines,
The copy being likely present other nodes and be cached with the second cache lines, so delays second when above-mentioned request bag
After depositing the operation of traveling row write, then the second cache lines just have modified.But simply to second in Home LLC
Cache lines is modified, and the second cache lines of other nodal caches is not revised, thus needs this
In LLC in a little nodes, in caching, the copy of the second cache lines is invalid.Such as: it is invalid to send out
(Invalid) information is by invalid for the copy of the second cache lines of caching in other nodes.Such as: as shown in Figure 4,
The node 401 at Home LLC place, wherein, the copy of the second cache lines is saved by distant-end node 402, far-end
Point 403 and distant-end node 404 LLC caching, so in this embodiment can by distant-end node 402,
The copy of the second cache lines of the LLC caching of distant-end node 403 and distant-end node 404 is invalid.
Furthermore it is also possible to be currency and described far-end enumerator current when described network-on-chip enumerator
When value sum is more than Preset Transfer overhead thresholds, by amended second cache lines from described Home LLC
When copying in the LLC of described requesting node and level cache, delay in described chip described the second of storage
Deposit capable copy invalid.Because after above-mentioned request bag carries out write operation to the second cache lines, then second delays
Deposit row just to have modified.But simply the second cache lines in Home LLC is modified, and other nodes
Second cache lines of caching is not the most revised, thus needs to delay in caching in the LLC in these nodes second
Deposit capable copy invalid.
In the present embodiment, on the basis of the embodiment shown in Fig. 1, add the embodiment of plurality of optional,
And can realize reducing the transport overhead in many-core framework.
Being apparatus of the present invention embodiment below, apparatus of the present invention embodiment is used for performing the inventive method embodiment
One to two method realized, for convenience of description, illustrate only the part relevant to the embodiment of the present invention, tool
Body technique details does not discloses, and refer to the embodiment of the present invention one and embodiment two.
Refer to the structural representation that Fig. 5, Fig. 5 are a kind of data distribution devices that the embodiment of the present invention provides,
As it is shown in figure 5, include: request unit 51, recognition unit 52 and the first copied cells 53, wherein:
Request unit 51, sends out for controlling the requesting node node to local Home last level cache LLC place
Send request bag.
Wherein, the node except above-mentioned Home LLC place during above-mentioned requesting node can be many-core framework it
Outer arbitrary node, above-mentioned Home LLC can be for data cached LLC in this many-core framework, its
In, the data of this Home LLC caching can be read by arbitrary node in this many-core framework, it addition, this crowd
Arbitrary node in core framework can also carry out write operation to the data of caching in Home LLC.
In the present embodiment, above-mentioned request bag can be read request packet, i.e. to Home LLC read operation；Or
Above-mentioned request bag can also is that write request bag, i.e. write operation in Home LLC.
Recognition unit 52, is used for when the node at described Home LLC place receives described request bag, root
The transport overhead between described requesting node and the node at described Home LLC place is identified according to described request bag
Whether exceed Preset Transfer overhead thresholds.
Above-mentioned transport overhead can be understood as the transmission road between the node at requesting node and Home LLC place
Footpath distance, or can be understood as the transmission path institute outside the node at requesting node and Home LLC place
The nodes of process；Or above-mentioned transport overhead it is also understood that and sends above-mentioned request bag extremely into requesting node
The duration that the node at Home LLC place is consumed, by the transport overhead of reflection request bag a length of time this.
First copied cells 53, for when described transport overhead exceedes Preset Transfer overhead thresholds, by described
The cache lines at the data place that request bag is asked copies to described requesting node from described Home LLC
In LLC.
In the first copied cells 53 when transport overhead exceedes Preset Transfer overhead thresholds, just explanation request joint
Point is relatively big, if requesting node reads in Home LLC with the node-node transmission expense at Home LLC place
During data, then the transport overhead of NoC will be very big, and step 103 copies data to requesting node
In LLC, so when requesting node uses these data, directly can read from the LLC of requesting node,
Such that it is able to the transport overhead reducing NoC will be the biggest；Wherein, this NoC may be used for the core of many-core framework
The network of multi-node communication on sheet.
Asked to read it addition, the cache lines at data place that above-mentioned request bag is asked can be request bag
The cache lines at data place of operation, or above-mentioned request bag asked to carry out the delaying of data place of write operation
Deposit row, and above-mentioned cache lines can be one or more cache lines.
Such as: the node 202 at requesting node 201 and Home LLC place, requesting node as shown in Figure 2
Transmission transmission path 203 as shown in Figure 2, path between the node 202 at 201 and Home LLC places.
As shown in Figure 2, between the node 202 at requesting node 201 and Home LLC place transmit path away from
From farther out, the transmission i.e. transmitting data between the node 202 at requesting node 201 and Home LLC place is opened
Pin can be bigger, and passes through above-mentioned steps by the data place required for cache request node in Home LLC
Cache lines copies to the LLC of requesting node 201, and when so next time uses these data, requesting node is just not required to
Send request bag, such that it is able to reduce transport overhead.
It addition, the cache lines at the data place asked by request bag in the first copied cells 53 is from described
Home LLC copies to the LLC of requesting node can be should according to the transmission path of transmission request bag
Cache lines copies to the LLC of requesting node；Or the data in the first copied cells 53, request bag asked
The cache lines at place copies to can be the data of chip in the LLC of requesting node from described Home LLC
This cache lines is copied directly in the LLC of requesting node by processor.It addition, above-mentioned duplication can be to replicate
The copy of cache lines, above-mentioned cache lines can be Cacheline, or cache line.
In the present embodiment, said method can apply in the chip of any many-core framework, or is applied to appoint
Meaning includes the smart machine of the chip of many-core framework.
In the present embodiment, the requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify described according to described request bag
Whether the transport overhead between the node at requesting node and described Home LLC place exceedes Preset Transfer expense
Threshold value；When described transport overhead exceedes Preset Transfer overhead thresholds, the data that described request bag is asked
The cache lines at place copies in the LLC of described requesting node from described Home LLC.Due to can be by
The cache lines at the data place of requesting node request is cached in the LLC of this requesting node, so this request joint
Point, when using these data, directly obtains just, without removing Home LLC from the LLC of this node
The node at place obtains, such that it is able to the transport overhead reduced in many-core framework.
Refer to the structural representation that Fig. 6, Fig. 6 are the another kind of data distribution devices that the embodiment of the present invention provides,
As shown in Figure 6, including: request unit 61, recognition unit 62 and the first copied cells 63, wherein:
Request unit 61, sends request bag for controlling the requesting node node to Home LLC place, its
In, the described requesting node node to Home LLC place is described during sending request bag asks bag every
In a node, described request bag, network-on-chip enumerator adds N, wherein, described N for more than or
Person's integer equal to 1.
Wherein, the requesting node node to Home LLC place send request bag be by requesting node with
Node-node transmission between the node at Home LLC place, transmission path 202 as shown in Figure 2.So should
Request bag often through a node, this node just by request wrap in network-on-chip enumerator add N (such as:
Add 1), when this request bag that the node at Home LLC place obtains, it is possible to by this request bag
This request bag of currency identification of network-on-chip counting have passed through many each and every one node, the i.e. transmission of this request bag and opens
Pin.It addition, when requesting node sends above-mentioned request bag, can be the network-on-chip counting in this request bag
Device sets an initial value, such as: 0.It addition, above-mentioned network-on-chip enumerator can be defined as (NoC Count,
NC), this NC is for adding up the transmission range that request wraps in network-on-chip, and this NC may be located at request bag
Packet header in.
Recognition unit 62, for when the node at described Home LLC place receives described request bag, sentencing
Whether the currency of the described network-on-chip enumerator in disconnected described request bag is more than Preset Transfer overhead thresholds.
Wherein, above-mentioned Preset Transfer overhead thresholds can be that user is set in advance, such as: 5,10,20 etc.,
Specifically can set according to the quantity of many-core framework interior joint, or according to the transmission path of many-core framework and
Set.
First copied cells 63, for the currency when network-on-chip enumerator more than Preset Transfer overhead thresholds,
The cache lines at the data place asked by described request bag copies to described request from described Home LLC
In the LLC of node.
When the currency of network-on-chip enumerator is more than Preset Transfer overhead thresholds, just explanation requesting node with
In Home LLC, between the node at place, transmission range is farther out.
In this enforcement just, above-mentioned request bag can be that the node to described Home LLC place reads data
Read request packet, i.e. requesting node read certain data to Home LLC request.Such as: the one-level of requesting node
After caching (L1Cache) sends disappearance (miss), and the LLC in requesting node does not read institute
During the data needed, requesting node just can send to the node at the Home LLC place being cached with these data
Read request packet.Wherein, requesting node can be by directory search to the data being cached with required for requesting node
Home LLC.
In this embodiment, described device can also include:
Counting unit 64, is used for when the node at described Home LLC place receives described request bag, will
The local enumerator being used for adding up the reading times that described requesting node reads the first cache lines preset adds N,
Wherein, the cache lines at the data place that read request packet described in described first caching behavior is read；
Described recognition unit 62 may be used for judging the current of the described described network-on-chip enumerator asked in bag
Whether value is more than Preset Transfer overhead thresholds with the currency sum of described local enumerator；
Described first copied cells 63 may be used for the currency when described network-on-chip enumerator and described this locality
When the currency sum of enumerator is more than described Preset Transfer overhead thresholds, by described first cache lines from described
Home LLC copies in LLC and the level cache of described requesting node.
Wherein, above-mentioned local enumerator is set in advance, and local enumerator is caching behavior granularity,
And can be that each cache lines sets multiple local enumerator, each local enumerator is for one joint of statistics
Point reads the number of times of this cache lines, such as: can set M local enumerator for above-mentioned first cache lines,
Wherein, a number of times reading the first cache lines for adding up above-mentioned requesting node deposited by M local enumerator.
It addition, above-mentioned local enumerator can be defined as (Home Count, HC), this HC is for adding up this locality
Cache lines is accessed for frequency.
This embodiment can realize add up requesting node and read the number of times of above-mentioned first cache lines, work as request
Node read the number of times of the first cache lines and request bag the number of nodes sum of process more than described Preset Transfer
During overhead thresholds, described first cache lines is copied to from described Home LLC the LLC of described requesting node
With in level cache.So can realize reading the first cache lines number of times with requesting node and request bag is passed through
Nodes the two dimension decide whether to copy to the first cache lines the LLC of requesting node.Because working as
Requesting node read the number of times of the first cache lines more time, illustrate that the first cache lines is heavier for requesting node
, such that it is able to the first cache lines copies to the LLC of requesting node, and requesting node is repeatedly from Home
LLC reads the first cache lines, thus causes bigger transport overhead；It addition, as requesting node and Home LLC
Between transmission path farther out time, requesting node reads the first cache lines from Home LLC and will also result in bigger
Transport overhead.
It addition, the first cache lines can be copied to by present embodiment in the level cache of requesting node, this
Sample requesting node can realize when level cache sends miss the data required for level cache are copied to one
In level caching, thus solve miss problem.
In the present embodiment, described device can also include:
Second copied cells 65, is used for the currency when described network-on-chip enumerator and described local enumerator
Currency sum less than or equal to described Preset Transfer overhead thresholds time, by described first cache lines from institute
State in the level cache copying to described requesting node in Home LLC.
This embodiment can realize requesting node and can realize level cache when level cache sends miss
Required data copy in level cache, thus solve miss problem.Further, since network-on-chip meter
The currency of number device is less than or equal to described Preset Transfer overhead thresholds with the currency sum of local enumerator
Time, illustrate that requesting node reads the number of times of the first cache lines less, or requesting node and Home LLC institute
Node between transmission path close together, it is possible to the first cache lines is not copied to request joint
In the LLC of point.
In the present embodiment, above-mentioned request bag can be that the second cache lines to described Home LLC carries out writing behaviour
The write request bag made；
Described recognition unit 62 may be used for judging the current of the described described network-on-chip enumerator asked in bag
Whether value is more than Preset Transfer overhead thresholds, wherein, described far-end meter with the currency sum of far-end enumerator
Number device is the reading times pre-setting and reading described second cache lines for adding up distant-end node, described far-end
Node is the node in chip in addition to the node at described Home LLC place；
Described first copied cells may be used for the currency when described network-on-chip enumerator and far-end enumerator
Currency sum more than Preset Transfer overhead thresholds time, by amended second cache lines from described Home
Copying in LLC and the level cache of described requesting node in LLC, wherein, described amended second delays
Deposit write request bag described in behavior response and described second cache lines is carried out write operation and the cache lines that generates.
Said distal ends node is all nodes in chip in addition to the node at described Home LLC place, or
Person's said distal ends node be in chip in addition to the node and above-mentioned requesting node at described Home LLC place
All nodes.So can count the second cache lines by said distal ends enumerator to be read or accessed
Number of times or frequency.It addition, said distal ends enumerator can be defined as (Remote Count, RC).Example
As: when distant-end node reads the second cache lines, this RC adds 1.
It addition, in this embodiment, request bag can also be carried out by above-mentioned request bag according to consistency protocol
The corresponding information of data of write operation be sent to the node at Home LLC place.
It addition, said apparatus can be also used for responding above-mentioned request bag, to the second cache lines in this embodiment
Carry out write operation.
This write operation can include amendment, add or the write operation such as deletion.By second after this write operation
Cache lines is above-mentioned amended second cache lines.
This embodiment can realize when network-on-chip enumerator currency and far-end enumerator currency it
With during more than Preset Transfer overhead thresholds, the cache lines of carried out for requesting node write operation is copied to request joint
In the LLC of point
In this embodiment, as it is shown in fig. 7, described device can also include:
Invalid unit 66, the copy of described second cache lines for storing in described chip is invalid, wherein,
The data of the second cache lines storage described in when the copy of described second cache lines is not respond described write request bag.
Such as: when network-on-chip enumerator currency and far-end enumerator currency sum less than or etc.
When Preset Transfer overhead thresholds, amended second cache lines is not copied to from described Home LLC
In the LLC of described requesting node and level cache, simply by described second cache lines of storage in described chip
Copy invalid.Because in the chips in addition to above-mentioned Home LLC is buffered in the copy of the second cache lines,
The copy being likely present other nodes and be cached with the second cache lines, so delays second when above-mentioned request bag
After depositing the operation of traveling row write, then the second cache lines just have modified.But simply to second in Home LLC
Cache lines is modified, and the second cache lines of other nodal caches is not revised, thus needs this
In LLC in a little nodes, in caching, the copy of the second cache lines is invalid.Such as: it is invalid to send out
(Invalid) information is by invalid for the copy of the second cache lines of caching in other nodes.Such as: as shown in Figure 4,
The node 401 at Home LLC place, wherein, the copy of the second cache lines is saved by distant-end node 402, far-end
Point 403 and distant-end node 404 LLC caching, so in this embodiment can by distant-end node 402,
The copy of the second cache lines of the LLC caching of distant-end node 403 and distant-end node 404 is invalid.
Furthermore it is also possible to be currency and described far-end enumerator current when described network-on-chip enumerator
When value sum is more than Preset Transfer overhead thresholds, by amended second cache lines from described Home LLC
When copying in the LLC of described requesting node and level cache, delay in described chip described the second of storage
Deposit capable copy invalid.Because after above-mentioned request bag carries out write operation to the second cache lines, then second delays
Deposit row just to have modified.But simply the second cache lines in Home LLC is modified, and other nodes
Second cache lines of caching is not the most revised, thus needs to delay in caching in the LLC in these nodes second
Deposit capable copy invalid.
In the present embodiment, on the basis of the embodiment shown in Fig. 5, add the embodiment of plurality of optional,
And can realize reducing the transport overhead in many-core framework.
Refer to the chip structure schematic diagram that Fig. 8, Fig. 8 are a kind of many-core frameworks that the embodiment of the present invention provides,
As shown in Figure 8, including: M node 81, communication bus 82, processor 83 and memorizer 84, wherein,
Wherein, described M node includes the node 812 at requesting node 811 and Home LLC place, wherein,
Requesting node 811 includes that level cache and LLC, the node 812 at Home LLC place include Home LLC,
Wherein, communication bus 82 is used for connecting each node 81, processor 83 and memorizer 84, wherein, described
Processor 83 performs to operate as follows for the program calling memorizer 84 storage:
Control the requesting node 811 node 812 to Home LLC place and send request bag；
When the node 812 at described Home LLC place receives described request bag, according to described request bag
Identify whether the transport overhead between the node at described requesting node and described Home LLC place exceedes default
Transport overhead threshold value；
When described transport overhead exceedes Preset Transfer overhead thresholds, the data institute that described request bag is asked
Cache lines copy to from described Home LLC in the LLC of described requesting node 811.
In the present embodiment, the requesting node node to Home LLC place is described during sending request bag
Request bag often in a node, described request bag network-on-chip enumerator can add N, wherein, institute
Stating N is the integer more than or equal to 1；
What processor 83 performed identifies described requesting node and described Home LLC place according to described request bag
Node between transport overhead whether exceed the operation of Preset Transfer overhead thresholds, may include that
Judge whether the currency of the described network-on-chip enumerator in described request bag is more than Preset Transfer expense
Threshold value.
In the present embodiment, request bag can be that the reading of the node reading data to described Home LLC place please
Seek bag；
The operation that processor 83 performs can also include:
When the node at described Home LLC place receives described request bag, default being used for is added up institute
The local enumerator stating the reading times that requesting node reads the first cache lines adds N, and wherein, described first delays
Deposit the cache lines at the data place that read request packet described in behavior is read；
What processor 83 performed judges that the currency of the described network-on-chip enumerator in described request bag is the biggest
In the operation of Preset Transfer overhead thresholds, may include that
Judge the currency of described network-on-chip enumerator in described request bag and working as of described local enumerator
Whether front value sum is more than Preset Transfer overhead thresholds；
Processor 83 perform when described transport overhead exceedes Preset Transfer overhead thresholds, by described request bag
The cache lines at the data place asked copies in the LLC of described requesting node from described Home LLC
Operation, may include that
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator are more than described
During Preset Transfer overhead thresholds, described first cache lines is copied to described request from described Home LLC
In the LLC of node and level cache.
In the present embodiment, the operation that processor 83 performs can also include:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator less than or
During equal to described Preset Transfer overhead thresholds, described first cache lines is copied to from described Home LLC
In the level cache of described requesting node.
In the present embodiment, request bag can be that the second cache lines to described Home LLC carries out write operation
Write request bag；
What processor 83 performed judges that the currency of the described network-on-chip enumerator in described request bag is the biggest
In the operation of Preset Transfer overhead thresholds, may include that
Judge the currency of the described described network-on-chip enumerator asked in bag and the currency of far-end enumerator
Whether sum is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is for pre-setting for adding up
Distant-end node reads the reading times of described second cache lines, and described distant-end node is except described Home in chip
Node outside the node at LLC place；
Processor 83 perform when described transport overhead exceedes Preset Transfer overhead thresholds, by described request bag
The cache lines at the data place asked copies in the LLC of described requesting node from described Home LLC
Operation, may include that
When the currency of described network-on-chip enumerator and the currency sum of described far-end enumerator are more than presetting
During transport overhead threshold value, amended second cache lines is copied to described request from described Home LLC
In the LLC of node and level cache, wherein, write request described in described amended second caching behavior response
The cache lines that bag carries out write operation to described second cache lines and generates.
In the present embodiment, the operation that processor 83 performs can also include:
By invalid for the copy of described second cache lines of storage in described chip, wherein, described second cache lines
Copy data of the second cache lines storage described in when being not respond described write request bag.
In the present embodiment, the requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify described according to described request bag
Whether the transport overhead between the node at requesting node and described Home LLC place exceedes Preset Transfer expense
Threshold value；When described transport overhead exceedes Preset Transfer overhead thresholds, the data that described request bag is asked
The cache lines at place copies in the LLC of described requesting node from described Home LLC.Due to can be by
The cache lines at the data place of requesting node request is cached in the LLC of this requesting node, so this request joint
Point, when using these data, directly obtains just, without removing Home LLC from the LLC of this node
The node at place obtains, such that it is able to the transport overhead reduced in many-core framework.
One of ordinary skill in the art will appreciate that all or part of flow process realizing in above-described embodiment method,
Can be by computer program and complete to instruct relevant hardware, described program can be stored in a calculating
In machine read/write memory medium, this program is upon execution, it may include such as the flow process of the embodiment of above-mentioned each method.
Wherein, described storage medium can be magnetic disc, CD, read-only store-memory body (Read-Only Memory,
Or random access memory (Random Access Memory, be called for short RAM) etc. ROM).
Above disclosed be only present pre-ferred embodiments, certainly can not with this limit the present invention it
Interest field, the equivalent variations therefore made according to the claims in the present invention, still belong to the scope that the present invention is contained.
Claims (12)
1. a data distributing method, it is characterised in that including:
The requesting node node to local Home last level cache LLC place sends request bag；
When the node at described Home LLC place receives described request bag, identify according to described request bag
Whether the transport overhead between the node at described requesting node and described Home LLC place exceedes Preset Transfer
Overhead thresholds；
When described transport overhead exceedes Preset Transfer overhead thresholds, the data institute that described request bag is asked
Cache lines copy to from described Home LLC in the LLC of described requesting node.
2. the method for claim 1, it is characterised in that described requesting node is to Home LLC institute
Node send request bag during the described request every sheet in a node, described request bag of bag
Upper network counter adds N, and wherein, described N is the integer more than or equal to 1；
Described according between the described request bag described requesting node of identification and the node at described Home LLC place
Transport overhead whether exceed Preset Transfer overhead thresholds, including:
Judge whether the currency of the described network-on-chip enumerator in described request bag is more than Preset Transfer expense
Threshold value.
3. method as claimed in claim 2, it is characterised in that described request bag is to described Home LLC
The node at place reads the read request packet of data；
Described method also includes:
When the node at described Home LLC place receives described request bag, default being used for is added up institute
The local enumerator stating the reading times that requesting node reads the first cache lines adds N, and wherein, described first delays
Deposit the cache lines at the data place that read request packet described in behavior is read；
Whether the currency of the described described network-on-chip enumerator judged in described request bag is more than Preset Transfer
Overhead thresholds, including:
Judge the currency of described network-on-chip enumerator in described request bag and working as of described local enumerator
Whether front value sum is more than Preset Transfer overhead thresholds；
Described when described transport overhead exceedes Preset Transfer overhead thresholds, that described request bag is asked number
Copy in the LLC of described requesting node from described Home LLC according to the cache lines at place, including:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator are more than described
During Preset Transfer overhead thresholds, described first cache lines is copied to described request from described Home LLC
In the LLC of node and level cache.
4. method as claimed in claim 3, it is characterised in that described method also includes:
When the currency of described network-on-chip enumerator and the currency sum of described local enumerator less than or
During equal to described Preset Transfer overhead thresholds, described first cache lines is copied to from described Home LLC
In the level cache of described requesting node.
5. method as claimed in claim 2, it is characterised in that described request bag is to described Home LLC
The second cache lines carry out the write request bag of write operation；
Whether the currency of the described described network-on-chip enumerator judged in described request bag is more than Preset Transfer
Overhead thresholds, including:
Judge the currency of the described described network-on-chip enumerator asked in bag and the currency of far-end enumerator
Whether sum is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is for pre-setting for adding up
Distant-end node reads the reading times of described second cache lines, and described distant-end node is except described Home in chip
Node outside the node at LLC place；
Described when described transport overhead exceedes Preset Transfer overhead thresholds, that described request bag is asked number
Copy in the LLC of described requesting node from described Home LLC according to the cache lines at place, including:
When the currency of described network-on-chip enumerator and the currency sum of described far-end enumerator are more than presetting
During transport overhead threshold value, amended second cache lines is copied to described request from described Home LLC
In the LLC of node and level cache, wherein, write request described in described amended second caching behavior response
The cache lines that bag carries out write operation to described second cache lines and generates.
6. method as claimed in claim 5, it is characterised in that described method also includes:
By invalid for the copy of described second cache lines of storage in described chip, wherein, described second cache lines
Copy data of the second cache lines storage described in when being not respond described write request bag.
7. a data distribution device, it is characterised in that including: request unit, recognition unit and first are multiple
Unit processed, wherein:
Described request unit, for controlling the requesting node node to local Home last level cache LLC place
Send request bag；
Described recognition unit, is used for when the node at described Home LLC place receives described request bag,
Identify that the transmission between described requesting node and the node at described Home LLC place is opened according to described request bag
Whether pin exceedes Preset Transfer overhead thresholds；
Described first copied cells, for when described transport overhead exceedes Preset Transfer overhead thresholds, by institute
The cache lines stating the data place that request bag is asked copies to described requesting node from described Home LLC
LLC in.
8. device as claimed in claim 7, it is characterised in that described requesting node is to Home LLC institute
Node send request bag during the described request every sheet in a node, described request bag of bag
Upper network counter adds N, and wherein, described N is the integer more than or equal to 1；
Whether described recognition unit for judging the currency of the described network-on-chip enumerator in described request bag
More than Preset Transfer overhead thresholds.
9. device as claimed in claim 8, it is characterised in that described request bag is to described Home LLC
The node at place reads the read request packet of data；
Described device also includes:
Counting unit, for when the node at described Home LLC place receives described request bag, by pre-
If add N for the local enumerator adding up the reading times that described requesting node reads the first cache lines, its
In, the cache lines at the data place that read request packet described in described first caching behavior is read；
Described recognition unit is for judging currency and the institute of the described network-on-chip enumerator in described request bag
Whether state the currency sum of local enumerator more than Preset Transfer overhead thresholds；
Described first copied cells is used for the currency when described network-on-chip enumerator and described local enumerator
Currency sum more than described Preset Transfer overhead thresholds time, by described first cache lines from described Home
LLC copies in LLC and the level cache of described requesting node.
10. device as claimed in claim 9, it is characterised in that described device also includes:
Second copied cells, is used for the currency when described network-on-chip enumerator and described local enumerator
When currency sum is less than or equal to described Preset Transfer overhead thresholds, by described first cache lines from described
Home LLC copies in the level cache of described requesting node.
11. devices as claimed in claim 8, it is characterised in that described request bag is to described Home LLC
The second cache lines carry out the write request bag of write operation；
Described recognition unit is for judging that the currency of the described network-on-chip enumerator in described request bag is with remote
Whether the currency sum of end enumerator is more than Preset Transfer overhead thresholds, and wherein, described far-end enumerator is
Pre-setting the reading times reading described second cache lines for adding up distant-end node, described distant-end node is
Node in addition to the node at described Home LLC place in chip；
Described first copied cells is used for when the currency of described network-on-chip enumerator and working as of far-end enumerator
When front value sum is more than Preset Transfer overhead thresholds, by amended second cache lines from described Home LLC
In copy in LLC and the level cache of described requesting node, wherein, described amended second cache lines
The cache lines generated for described second cache lines is carried out the described write request bag of response write operation.
12. devices as claimed in claim 11, it is characterised in that described device also includes:
Invalid unit, the copy of described second cache lines for storing in described chip is invalid, wherein,
The data of the second cache lines storage described in when the copy of described second cache lines is not respond described write request bag.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510317846.2A CN106302614A (en) | 2015-06-11 | 2015-06-11 | A kind of data distributing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510317846.2A CN106302614A (en) | 2015-06-11 | 2015-06-11 | A kind of data distributing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106302614A true CN106302614A (en) | 2017-01-04 |
Family
ID=57660300
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510317846.2A Pending CN106302614A (en) | 2015-06-11 | 2015-06-11 | A kind of data distributing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106302614A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060143408A1 (en) * | 2004-12-29 | 2006-06-29 | Sistla Krishnakanth V | Efficient usage of last level caches in a MCMP system using application level configuration |
| CN102117262A (en) * | 2010-12-21 | 2011-07-06 | 清华大学 | Method and system for active replication for Cache of multi-core processor |
| CN203340113U (en) * | 2013-06-28 | 2013-12-11 | 环境保护部华南环境科学研究所 | Source server system for through intranet-extranet long-distance data transmission and target server system |
| CN103744799A (en) * | 2013-12-26 | 2014-04-23 | 华为技术有限公司 | Memory data access method, device and system |
| CN104679688A (en) * | 2013-12-02 | 2015-06-03 | 华为技术有限公司 | Data access method, device and system |
-
2015
- 2015-06-11 CN CN201510317846.2A patent/CN106302614A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060143408A1 (en) * | 2004-12-29 | 2006-06-29 | Sistla Krishnakanth V | Efficient usage of last level caches in a MCMP system using application level configuration |
| CN102117262A (en) * | 2010-12-21 | 2011-07-06 | 清华大学 | Method and system for active replication for Cache of multi-core processor |
| CN203340113U (en) * | 2013-06-28 | 2013-12-11 | 环境保护部华南环境科学研究所 | Source server system for through intranet-extranet long-distance data transmission and target server system |
| CN104679688A (en) * | 2013-12-02 | 2015-06-03 | 华为技术有限公司 | Data access method, device and system |
| CN103744799A (en) * | 2013-12-26 | 2014-04-23 | 华为技术有限公司 | Memory data access method, device and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104809076B (en) | Cache management method and device | |
| EP1667026A2 (en) | Managing replacement of data in a cache on a node based on caches of other nodes | |
| CN105701219B (en) | A kind of implementation method of distributed caching | |
| CN106027595A (en) | Access log processing method and system for CDN node | |
| CN101662483A (en) | Cache system for cloud computing system and method thereof | |
| CN107491320A (en) | Loading method and device based on mixed mode Mobile solution | |
| CN110134896A (en) | A kind of the monitoring process and intelligent buffer method of proxy server | |
| CN107864215B (en) | Peer-to-peer network file system, access control/management method/system, and terminal | |
| CN107341114A (en) | A kind of method of directory management, Node Controller and system | |
| CN107370809A (en) | Method of data synchronization and data search system | |
| US6647469B1 (en) | Using read current transactions for improved performance in directory-based coherent I/O systems | |
| CN110244909A (en) | To contain the device and method of the distributed memory systems of memory node | |
| CN109359063A (en) | Caching replacement method, storage equipment and storage medium towards storage system software | |
| CN107506154A (en) | A kind of read method of metadata, device and computer-readable recording medium | |
| CN109327506A (en) | A kind of resource allocation methods, device and readable storage medium storing program for executing | |
| CN106250322B (en) | A kind of method and apparatus for writing data | |
| CN107633102A (en) | A kind of method, apparatus, system and equipment for reading metadata | |
| CN107197000A (en) | Static dynamic hybrid cache method, apparatus and system | |
| CN105471955A (en) | Writing method of distributed file system, client device and distributed file system | |
| CN103825916A (en) | Resource downloading method and resource downloading system | |
| CN106713460A (en) | Version control method and system for flash application | |
| CN108459821A (en) | A kind of method and device of data buffer storage | |
| CN106506640B (en) | A kind of processing method and processing device of SNMP request | |
| CN106302614A (en) | A kind of data distributing method and device | |
| CN106919338B (en) | A kind of Flash storage access processing method and system suitable for software definition storage |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170104 |