CN102904829A - Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information - Google Patents
Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information Download PDFInfo
- Publication number
- CN102904829A CN102904829A CN2012103994043A CN201210399404A CN102904829A CN 102904829 A CN102904829 A CN 102904829A CN 2012103994043 A CN2012103994043 A CN 2012103994043A CN 201210399404 A CN201210399404 A CN 201210399404A CN 102904829 A CN102904829 A CN 102904829A
- Authority
- CN
- China
- Prior art keywords
- fast
- queuing delay
- algorithm
- delay
- maximum
- 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
Images
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention provides a unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information according to the characteristics of FAST TCP unilateral acceleration application specific to the open problems of difficulty in selecting protocol parameters and accurate estimation of time propagation delay existing in FAST TCP. The method comprises the following steps of: periodically collecting history information such as maximum queuing delay and starting time in each lower layer active FAST TCP; fully utilizing information such as maximum queuing delay which can be provided in history connection by using an upper layer algorithm; periodically and adaptively selecting protocol parameters on a large time scale; accurately estimating time propagation delay; and informing a lower layer of running FAST TCP connection. Due to the adoption of the method provided by the invention, the stability, utilization ratio and justification of a high-speed network FAST TCP system are improved greatly.
Description
Technical field
The present invention relates to a kind of express network transmission control protocol algorithm and improve the field, is a kind of raising high speed specifically
The monolateral acceleration FAST TCP based on historical connection information of the stability of the network transmission protocol, utilance and fairness changes
Advance algorithm.
Background technology
FAST TCP (Fast Active queue management Scalable Transmission Control Protocol is called for short FAST) is a kind of novel transmission control protocol that proposes for Next generation high-speed networks; In high speed network environment, lose with group-based and to improve Transmission Control Protocol as other of congestion feedback signal and compare, the FAST agreement adopts the information calculations FAST of queuing delay that estimates to connect the grouping number of staying the bottleneck link buffering area, stay the number of bottleneck link buffering area grouping as balance point take expectation, according to the actual grouping number of bottleneck link buffering area of staying apart from the distance of balance point position, non-linearly adjust the speed of send window size variation, do not need the network layer intermediate node to participate in, ACTIVE CONTROL is stayed the bottleneck link buffer queue length, thereby initiatively avoided the appearance of buffering area queue overflow and congestion phenomenon, obtained better stability and bottleneck link service efficiency more fully, its bandwidth availability ratio can reach more than 90%; But there is the suitable protocol parameter of How to choose in it and is difficult to quick and precisely estimate these two open problems of propagation delay; The open problem of these two keys has obstructed the FAST agreement in further application and popularization of express network just.
FAST system ACTIVE CONTROL is stayed bottleneck link buffer queue length and FAST protocol parameter
With active FAST linking number
Relevant; Protocol parameter
Refer to that FAST connects expectation and stays bottleneck link end buffer data grouping number; Relatively more commonly used is to adopt the static mappings table to select disposable specified protocol parameter according to the bottleneck link bandwidth at present
; But along with the continuous foundation of FAST connection,
Can constantly increase, these active connection ACTIVE CONTROL are stayed the bottleneck link buffer queue length
Also can constantly increase, when ACTIVE CONTROL is stayed the bottleneck link buffer queue length
Surpass the bottleneck link buffer memory capacity
The time, will produce and overflow; When this situation occurring, moving in circles of cache overflow, loss recovery, three kinds of states of burst rate can appear in the FAST system, finally causes serious message segment to lose and hang down QoS.
But owing to can't obtain to pass through accurately the FAST linking number of bottleneck link in source
With the bottleneck link buffer memory capacity
, therefore be difficult to select the proper protocol parameter
; The people such as Tang A have proposed under the number of minutes magnitude yardstick, adjust protocol parameter according to queuing delay and average packet loss rate dynamic adjustment
Strategy; This strategy focus be placed between agreement justice but not on the agreement internal fairness; Zhu Xiaosong has proposed according to current
Value and predicted value
Between difference come dynamically-adjusting parameter
Strategy, but should not provide how to choose predicted value by strategy
Method; Heying Zhang has proposed dynamically to adjust protocol parameter according to target queuing delay
AFAST TCP algorithm, but algorithm does not also specifically provide the method for how to confirm target queuing delay; Song Lihua notices that depending merely on FAST is to be difficult to solve selection proper protocol parameter in source
This open problem; Therefore having proposed increases one deck towards the centralized guidance function of trunk bottleneck link at the FAST source end system, periodically (as tens of second or several minutes) measure the performance index such as bottleneck bandwidth, round-trip delay of network backbone bottleneck link, according to results presumption competition number of connection, then calculate the number of redundant packets that each connection is got by expected length of the waiting line, notice respective ends system; Then undated parameter setting of end system; Song Lihua improves above-mentioned method according to measurement result supposition competition number of connection, adopts fuzzy control technology to instruct the FAST source to select suitable protocol parameter
; Song Lihua has proposed a kind of transport layer solution of 2 layers of structure further with the said method arrangement, utilizes the performance service-aware internal network state on upper strata, then forms on this basis the lower floor's control target based on redundancy packets; In sum, though the people such as Zhu Xiaosong propose corresponding improvement project, when in fact just will be to protocol parameter
Setting transfer to the expectation
The setting of value and expectation queuing delay does not fundamentally solve protocol parameter is set
This open problem, and because each connection can't direct communication, each connects can't adjust protocol parameter synchronously, exists protocol parameter to adjust the slow problem of convergence rate; Although Song Lihua has solved the disclosure problem, need to introduce the methods such as externally measured technology, there is deployment issue.
FAST connects minimum round trip delay time that source is obtained as propagation delay, thereby when network is in Persistent Congestion, be that bottleneck link has existed when being under the poised state FAST and connecting, newly-built FAST connects and will be difficult to fast, accurately estimates propagation delay, thereby causes new, old connection to show for a long time serious unjustness; Therefore, when network is in Persistent Congestion, newly-built FAST is connected obtain fast accurately propagation delay, reveal serious unjustness thereby solve new, old connection table, also is the key issue that the FAST agreement need to solve.
[0006] a kind of each FAST of improvement of Tan Liansheng proposition connects first bag queuing priority algorithm, this algorithm connects first bag that sends with each high priority is set, directly do not send so that this bag does not need to participate in queuing during through bottleneck link, thereby obtain the accurately propagation delay of this connection according to the RTT of first bag; This algorithm can obtain accurately propagation delay and the fairness that guarantees the FAST agreement really, but has the problem of deployment aspect; Tony Cui has pointed out that under the help that does not have the network equipments such as router the queue queue's (being called for short later on formation) that empties in short-term bottleneck link end cushion space is the unique method that the FAST source obtains accurate propagation delay; Therefore proposed each newly-established connection, take temporary transient rollback strategy after arriving balance point, this algorithm has reduced the stability of a system and utilance; Migule R has proposed a kind of improvement algorithm that utilizes the source local message accurately to estimate propagation delay, but the method is only effective when connecting to newly setting up one, if arrive simultaneously a plurality of new connections, just can't utilize local message to calculate old linking number and bottleneck link delivery flow rate, can't obtain accurately propagation delay; And owing to when improving new connection transmission rate, can't guarantee simultaneously that old connection does not respond, thereby also there is certain error in this algorithm itself.
In sum as can be known, also there is not good way to solve these two the open problems that the FAST agreement exists at present; Therefore, wish to find the method that solves the disclosure problem according to the commercial actual conditions of using of FAST; Be found in the FastSoft company of California, USA in 2006, released the monolateral expedite product FastSoft E of wide area network Series, the breakthrough network optimization technology Fast algorithm that this product has adopted California Institute of Technology to develop in 2004.The present invention is according to the practical application of the monolateral acceleration of above-mentioned FAST, a kind of transport layer solution of 2 layers of structure has been proposed, lower floor still uses the FAST algorithm, but each connect to need increase maximum queuing delay and start-up time 2 parameters, and hour between these 2 variablees of yardstick periodic maintenance; Notice upper strata algorithm when packet loss occurs; After the upper strata algorithm finds that there is the packet loss of connection in lower floor, read each active maximum queuing delay that connects of lower floor, determine target queuing delay control range according to this maximum queuing delay, every the number of minutes magnitude yardstick update cycle, self adaptation is adjusted protocol parameter, and notice connects to the FAST of lower floor's operation, and ACTIVE CONTROL bottleneck link buffer queue length is in rational scope, prevent bottleneck link buffering area packet loss, Systems balanth and utilance are provided; When newly-built connection was arranged, the upper strata algorithm did not need externally measured system to participate in according to the start-up time of history connection queuing delay the earliest, accurately estimates propagation delay; Because the active connection of the unified notice of upper strata algorithm lower floor has also solved each connection protocol parameter and can't communicate by letter the problem that can not restrain synchronously.
Summary of the invention
For two that not yet thoroughly solve in the present FAST algorithm open problems, should according to the reality of the monolateral acceleration of FAST
With, the present invention proposes a kind of monolateral acceleration FAST TCP based on historical connection information and improve algorithm.
Monolateral acceleration FAST TCP based on historical connection information provided by the invention improves algorithm, and lower floor's algorithm still adopts the FAST algorithm, but periodic maintenance originally connects the information such as maximum queuing delay, start-up time, notice upper strata algorithm when packet loss occurs; Specifically comprise the steps:
Step 1: the maximum queuing delay that in small scale interval cycle time is added on original algorithm basis, should connect
, this connects start-up time
, running time
Deng global variable;
Step 2: at the beginning of the small scale update cycle, arrange
=0, because onlying demand the maximum queuing delay in this cycle;
Step 3: whenever receive an acknowledgement frame, record it in the queuing delay of the maximum in this cycle, namely
Step 4: if (receive three identical acknowledgement frames, namely find to have the packet loss phenomenon) notice upper strata algorithm, bottleneck link generation packet loss;
To transmitting terminal
iIndividual FAST connection is defined as follows variable:
: transmitting terminal congestion window size (packets);
: propagation delay (s);
: queuing delay (s);
: the
iThe contact time delay that connects,
(s);
: average round trip delay time (s);
: delivery flow rate (packets/s), wherein
;
: protocol parameter (packets);
: the control law gain parameter;
: the transmitting terminal window update cycle (s);
: the upper strata algorithm small scale update cycle;
: the upper strata algorithm large scale update cycle,
Be bottleneck link
Bandwidth (packets/s);
The upper strata algorithm is made following response according to the packet loss notice that lower floor's algorithm provides, and specifically comprises the steps:
Step 1: if lower floor does not have FAST to connect packet loss, then do not start the upper-layer protocol adjustment algorithm; If receive the packet loss signal that lower floor has connection to send, then carry out following steps:
[0019] step 4: determine target queuing delay control range:
Step 5: call dynamic adjustment protocol parameter algorithm.
According to the packet drop of the lower floor's algorithm dynamic state of parameters adjustment algorithm that periodically carries on an agreement, specifically comprise the steps:
Step 2: if this algorithm triggers because of packet loss, then reclocking, until behind one-period, just can carry out following steps, to keep the number of minutes magnitude yardstick of adjustment;
Step 3: before had to connect packet loss occurs, if obtained the control range of maximum queuing delay and target queuing delay, if namely
, then record the group time-delay in last cycle
;
=0;
Step 5: the maximum queuing delay that obtains this update cycle
If, record the most famous time-delay of current period greater than maximum queuing delay, then replace it.Be If
, then
;
Step 7: if always the most maximum queuing delay does not then adjust, namely in the control target zone
If (
And
), then { need not take measures the target queuing delay's scope that met the expectation; }
Step 8: if the most maximum total queuing delay is then definite from expected range far and near position, i.e. If(less than the minimum value of control target zone
) then
;
Step 10: if the most maximum total queuing delay is then definite from expected range distance position greater than the maximum of control target zone, namely if (
), then
;
Step 11: determine target queuing delay according to above-mentioned position distance: ';
Step 12: according to target queuing delay, determine adjusting range:
Step 13: self adaptation is adjusted protocol parameter:
Step 14: modifying agreements parameter area:
A kind of connection for new FAST is difficult to accurately the fast improvement algorithm of acquisition propagation delay, it is characterized in that, specifically comprises the steps:
Step 1: the yardstick cycle is read the active current time queuing delay that connects between each hour
, settling time
And running time;
;
Step 2: the FAST of selection foundation the earliest connects from active connection
Queuing delay as bottleneck link
;
Step 3: when having newly-built connection to set up, provide the current queuing delay that connects with the active FAST that sets up the earliest in history of the identical destination host of this newly-built connection
;
Step 4: to the FAST of lower floor algorithm, when having newly-built FAST connection to set up, revise the method that it calculates propagation delay for the first time:
Each new FAST is connected:
The present invention's characteristics that monolateral acceleration is used according to FAST, the transport layer that has proposed a kind of 2 layers of structure is improved algorithm; Each active FAST of lower floor connects the historical informations such as the maximum queuing delay of regularly collection, start-up time, the upper strata algorithm takes full advantage of the information such as maximum queuing delay that historical connection can provide, large time scale periodic time self-adapting selection protocol parameter, accurately estimate propagation delay, notice connects to the FAST of lower floor's operation.
Case verification this improvement algorithm significantly improved Systems balanth, utilance and fairness.
Description of drawings
The monolateral expedite product FastSoft E of Fig. 1 Series uses.
The monolateral accelerating system network topology of Fig. 2 FAST.
Fig. 3 (a) is two layers of total improvement algorithm, (b) be module 1 among Fig. 3 (a), 2 flow chart, (c) be that the flow chart (d) of the response packet loss algorithm of module 3 among Fig. 3 (a) is the flow chart of the dynamic adjustment protocol parameter of module 4 among Fig. 3 (a), (e) improve the flow chart of FAST fairness algorithm.
Fig. 4 is for adopting former FAST algorithmic system operational effect, and wherein (a) is bottleneck link buffering area packet loss number, (b) is the bottleneck link buffer queue length, (c) is the delivery flow rate that three paths distribute.
Fig. 5 improves the algorithmic system operational effect for adopting based on two layers of FAST, and wherein (a) is bottleneck link buffering area packet loss number, (b) is the bottleneck link buffer queue length, (c) is the delivery flow rate that three paths distribute.
Fig. 6 is in the link section buffer capacity situation that can change, employing improves the algorithmic system operational effect based on two layers of FAST, wherein (a) is bottleneck link buffering area packet loss number, (b) is the bottleneck link buffer queue length, (c) is the delivery flow rate that three paths distribute.
Fig. 7 is for adopting the present invention to improve the running effect of FAST fairness algorithm and employing traditional F AST algorithm, and wherein (a) is the delivery flow rate that three paths distribute for adopting traditional algorithm; (b) improve the delivery flow rate that algorithm three paths distribute for the present invention.
Embodiment
FastSoft E Series product has adopted the patented technology FAST algorithm of California Institute of Technology, this product is mainly used in the monolateral acceleration from the server to client end as shown in Figure 1, as shown in Figure 1, as long as at a standard 1U server size of the simple series connection of server end FastSoft E Series equipment, need not newly-increased software or browser plug-in in client, and also need not the modification configuration or rewrite code to realize monolateral acceleration function at server end, access this server in arbitrary place in the whole world and all can enjoy 30% to 500% acceleration of dynamic page file transfer, main application fields: (1) SaaS (Software as a Service – part of Cloud computing); (2) Video/Gaming Acceleration; (3) File Transfer; (4) CDN (Content Delivery Network); The major customer comprises MySpace, Overstock; Com, Limelight Networks, JWT, Thomson Technicolor, Honda, Siemens etc.
According to the real network model of Fig. 1 as can be known, realize that just server arrives the monolateral acceleration function of each client, so only have an information source host node; The FAST that sets up in information source main frame and each stay of two nights main frame always has a bottleneck link in connecting; Therefore can make up network topological diagram as shown in Figure 2; In Fig. 2, suppose that S1 is the information source host node, D1, D2 ... D
MStay of two nights host node; Intermediate node L1, L2 form bottleneck link
l; Information source main frame, some phase connected links and stay of two nights main frame form a paths; Comprise information source main frame S1, link L1-L2 and stay of two nights main frame D1 such as path: S1-L1-L2-D1, every paths can be set up a plurality of FAST and connect.
The present invention makes up 2 layers of architecture advances algorithm as shown in Figure 3 on the basis of the employed FAST algorithm of FastSoft E Series equipment; Lower floor's algorithm still uses traditional FAST agreement, but each connection need to hour between the maximum queuing delay of yardstick periodic maintenance, when packet loss occurs, to notify immediately the upper strata algorithm; The main purpose of upper strata algorithm is every a large time scale update cycle, provides a suitable protocol parameter to the improved FAST source of lower floor
; Its design philosophy is: when initial condition, if do not connect packet loss, then directly adopt the static mappings table method to select protocol parameter; As find that lower floor has FAST to connect packet loss, then get interior all connections of previous small scale cycle of packet loss and measure the queuing delay of maximum bottleneck link as the maximum queuing delay of bottleneck link, the target queuing delay control range of expectation is set according to maximum queuing delay, bottleneck link queuing delay carries out the adjustment of protocol parameter every a large scale update cycle, so that can be controlled in the scope of expectation.
Lower floor's algorithm still adopts the FAST algorithm, but periodic maintenance originally connects the information such as maximum queuing delay, start-up time, and notice upper strata algorithm specifically comprises the steps: when packet loss occurs
Step 1: the maximum queuing delay that in small scale interval cycle time is added on original algorithm basis, should connect
, this connects start-up time
, running time
Deng global variable;
Step 2: at the beginning of the small scale update cycle, arrange
=0, because onlying demand the maximum queuing delay in this cycle;
Step 3: whenever receive an acknowledgement frame, record it in the queuing delay of the maximum in this cycle, namely
Step 4: if (receive three identical acknowledgement frames, namely find to have the packet loss phenomenon) notice upper strata algorithm, bottleneck link generation packet loss;
To transmitting terminal
iIndividual FAST connection is defined as follows variable:
: transmitting terminal congestion window size (packets);
: propagation delay (s);
: queuing delay (s);
: the
iThe contact time delay that connects,
(s);
: average round trip delay time (s);
: delivery flow rate (packets/s), wherein
;
: protocol parameter (packets);
: the control law gain parameter;
: the transmitting terminal window update cycle (s);
: the upper strata algorithm small scale update cycle;
: the upper strata algorithm large scale update cycle,
Be bottleneck link
Bandwidth (packets/s).
The upper strata algorithm is made following response according to the packet loss notice that lower floor's algorithm provides, and specifically comprises the steps:
Step 1: if lower floor does not have FAST to connect packet loss, then do not start the upper-layer protocol adjustment algorithm; If receive the packet loss signal that lower floor has connection to send, then carry out following steps:
Step 4: determine target queuing delay control range:
Step 5: call dynamic adjustment protocol parameter algorithm.
According to the packet drop of the lower floor's algorithm dynamic state of parameters adjustment algorithm that periodically carries on an agreement, specifically comprise the steps:
Step 1: per update cycle
Or because packet loss is carried out following steps;
Step 2: if this algorithm triggers because of packet loss, then reclocking, until behind one-period, just can carry out following steps, to keep the number of minutes magnitude yardstick of adjustment;
Step 3: before had to connect packet loss occurs, if obtained the control range of maximum queuing delay and target queuing delay, if namely
, then record the group time-delay in last cycle
;
=0;
Step 5: the maximum queuing delay that obtains this update cycle
If, record the most famous time-delay of current period greater than maximum queuing delay, then replace it.Be If
, then
;
Step 7: if always the most maximum queuing delay does not then adjust, namely in the control target zone
If (
And
), then { need not take measures the target queuing delay's scope that met the expectation; }
Step 8: if the most maximum total queuing delay is then definite from expected range far and near position, i.e. If(less than the minimum value of control target zone
) then
;
Step 10: if the most maximum total queuing delay is then definite from expected range distance position greater than the maximum of control target zone, namely if (
), then
;
Step 11: determine target queuing delay according to above-mentioned position distance:
Step 12: according to target queuing delay, determine adjusting range:
Step 13: self adaptation is adjusted protocol parameter:
; //
Step 14: modifying agreements parameter area:
A kind of connection for new FAST is difficult to accurately the fast improvement algorithm of acquisition propagation delay, specifically comprises the steps:
Step 1: the yardstick cycle is read the active current time queuing delay that connects between each hour
, settling time
And running time;
;
Step 2: the FAST of selection foundation the earliest connects from active connection
Queuing delay as bottleneck link
;
Step 3: when having newly-built connection to set up, provide the current queuing delay that connects with the active FAST that sets up the earliest in history of the identical destination host of this newly-built connection
;
Step 4: to the FAST of lower floor algorithm, when having newly-built FAST connection to set up, revise the method that it calculates propagation delay for the first time:
Each new FAST is connected:
Pass through NS2; 31 emulation come the validity of verification algorithm, and the network topology that emulation is adopted is supposed as shown in Figure 2
M=3,3 stay of two nights main frame D are namely arranged
1, D
2And D
3, consist of path S1-L1-L2-D1 (be called for short path S1-D1 behind the literary composition, the rest may be inferred by analogy for it), S1-L1-L2-D2 and S1-L1-L2-D3; Suppose that S1-L1, L2-D1, L2-D2 and L2-D3 have enough bandwidth and link buffer memory capacity, can congested and packet loss, but L2 to three stay of two nights main frame propagation delay is different, is respectively 10 (ms), 20 (ms) and 15 (ms); The bandwidth of bottleneck link L1-L2
(Mb/s), propagation delay is 50 (ms), the bottleneck link buffer memory capacity
=800 (packets); Suppose 1 (packets)=1000 (bytes)=8000 (bit), therefore
80 (Mb/s)=10000 (packets/s).
Bandwidth according to bottleneck link
(Mb/s), adopt the static mappings table method to determine each connection protocol parameter
; Simulation time 1000 (s), three paths are respectively set up 10 FAST and are connected; Path 1 connects when 0 (s), finishes when 1000 (s); Path 2 connects at 150 (s), and 600 (s) finish; Path 3 connects at 400 (s), and 1000 (s) finish; Suppose that each connection can both obtain accurately propagation delay, does not exist fairness problem.
Fig. 4 is adopting the results of running under the original FAST algorithm, as shown in Figure 4, at front 150 (s), when 10 that only have path S1-D1 active FAST connect,
; Therefore
, system can stablize; When 150 (s) and since the foundation of path S1-D2 10 FAST connect, at this moment
, therefore
, the bottleneck link buffer overflow begins to occur packet loss, and system begins to occur vibration; When to 400 (s), path S1-D3 has set up again 10 FAST and has connected,
, therefore, the bottleneck link buffering area is more crowded, and packet loss more aggravates row.
Fig. 5 adopts and the present invention is based on two layers of improvement of FAST algorithmic system operational effect, by the operation result of Fig. 5 as can be known, when 150 (s), because having set up 10 FAST, path S1-D2 connects, the bottleneck link buffer overflow, begin to occur packet loss, lower floor connects the discovery packet loss, can notify the upper strata algorithm at once; Packet loss information is received on the upper strata, starts dynamically to adjust protocol algorithm at once, adjusts corresponding protocol parameter
With the adjustment weight coefficient
, and in time notify to lower floor; Lower floor's algorithm carries out computing according to new protocol parameter; Comparison diagram 4, Fig. 5, owing to adopted dynamic protocol parameter adjustment algorithm, packet loss has only occured one time in each connection; , queuing delay can be controlled in the scope of expectation, each path justice divided the bottleneck link bandwidth equally, system presents good stability and fairness.
Fig. 6 changes in route, in the situation that namely the bottleneck link buffer size changes, adopts to the present invention is based on two layers of FAST and improve the algorithmic system operational effect, by the operation result of Fig. 6 as can be known, and when 200 (s), active linking number
=20, the bottleneck link buffering becomes 800 (packets) by 1200 (packets), therefore has
, the bottleneck link buffer overflow begins to occur packet loss, and self adaptation is adjusted protocol parameter
; When arriving 400 (s), there is again new connection to set up, during to 600 (s), the bottleneck link buffering further diminishes, and the situations such as 800 (s) Connection Release occur, but algorithm of the present invention can both the self adaptation adjustment; As shown in Figure 6, algorithm of the present invention can reduce the packet drop generation control of queue in suitable scope, and each path also can fair distribution bandwidth.
All enliven bottleneck link of FAST Connection Sharing, should have identical queuing delay, but FAST connects minimum round trip delay time that source is obtained as propagation delay, thereby exist when being in that FAST connects under the poised state when bottleneck link, newly-built FAST connects will be difficult to fast, accurately estimate propagation delay; But consider in the monolateral accelerating system network topology of FAST and only have an information source node that therefore some information that provide that can take full advantage of historical connection are provided newly-built FAST, quick and precisely obtain propagation delay; The operational effect of Fig. 7 (a) has illustrated this point.
Lower floor connects increases start-up time and operation time parameters, and each the connection when setting up, record connects start-up time and calculates running time; The upper strata algorithm is always in the information of safeguarding the active connection that starts the earliest; Connect when setting up when finding that lower floor has, can provide start-up time the earliest, the current queuing delay of the longest connection running time; The newly-built connection of lower floor changes the method for calculating propagation delay for the first time, and the queuing delay that deducts the upper strata algorithm and provide according to the current round trip delay time that measures calculates accurately propagation delay.Fig. 7 (b) is the operation result after employing the inventive method.Comparison diagram 7 (a) and (b) as can be known, 150 (s) with are connected (s) have respectively new connection to set up, the fairness when each path is greatly improved; When 600 (s), Connection Release is arranged, phenomenon has appearred emptying in short-term in formation, each connects and has obtained accurately propagation delay, fair distribution bandwidth.
The inventive method is tested under the different simulated environment such as each propagated time-delay difference, bottleneck link buffer queue change respectively; Experimental result shows, protocol parameter is dynamically adjusted in the variation that can initiatively conform of this improvement algorithm, accurately estimates propagation delay, and buffer queue length in the ACTIVE CONTROL bottleneck link improves utilance, stability and the fairness of bottleneck link.
Claims (4)
1. the monolateral acceleration FAST TCP based on historical connection information improves algorithm, lower floor's algorithm still adopts FAST TCP algorithm, but periodic maintenance originally connects the information such as maximum queuing delay, start-up time, notice upper strata algorithm when packet loss occurs, it is characterized in that, comprise the steps:
Step 1: the maximum queuing delay that in small scale interval cycle time is added on original algorithm basis, should connect
, this connects start-up time
, running time
Deng global variable;
Step 2: at the beginning of the small scale update cycle, arrange
=0, because onlying demand the maximum queuing delay in this cycle.
Step 3: whenever receive an acknowledgement frame, record it in the queuing delay of the maximum in this cycle, namely
Step 4: if receive three identical acknowledgement frames, namely find to have packet loss phenomenon notice upper strata algorithm, bottleneck link generation packet loss;
To transmitting terminal
iIndividual FAST TCP connection is defined as follows variable:
: transmitting terminal congestion window size (packets);
: propagation delay (s);
: queuing delay (s);
: the
iThe contact time delay that connects,
(s);
: average round trip delay time (s);
: delivery flow rate (packets/s), wherein
;
: protocol parameter (packets);
: the control law gain parameter;
: the transmitting terminal window update cycle (s);
: the small scale update cycle;
: the large scale update cycle,
Be bottleneck link
Bandwidth (packets/s).
2. the monolateral acceleration FAST TCP based on historical connection information according to claim 1 improves algorithm, it is characterized in that, the upper strata algorithm is made following response according to the packet loss notice that provides, and comprises the steps:
Step 1: if lower floor does not have FAST TCP to connect packet loss, then do not start the upper-layer protocol adjustment algorithm; If receive the packet loss signal that lower floor has connection to send, then carry out following steps:
Step 4: determine target queuing delay control range:
Wherein
;
Step 5: call dynamic adjustment protocol parameter algorithm.
3. the monolateral acceleration FAST TCP based on historical connection information according to claim 1 improves algorithm, it is characterized in that, calls dynamic adjustment protocol parameter algorithm and comprises the steps:
Step 1: per update cycle
Or because packet loss is carried out following steps;
Step 2: if this algorithm triggers because of packet loss, then reclocking, until behind one-period, just can carry out following steps, to keep the number of minutes magnitude yardstick of adjustment;
Step 3: before had to connect packet loss occurs, if obtained the control range of maximum queuing delay and target queuing delay, if namely
, then record the group time-delay in last cycle
;
=0;
Step 7: if always the most maximum queuing delay does not then adjust, namely in the control target zone
If (
And
), then { need not take measures the target queuing delay's scope that met the expectation; }
Step 8: if the most maximum total queuing delay is then definite from expected range far and near position, i.e. If(less than the minimum value of control target zone
) then
Step 10: if the most maximum total queuing delay is then definite from expected range distance position greater than the maximum of control target zone, namely if (
), then
Step 11: determine target queuing delay according to above-mentioned position distance:
Step 12: according to target queuing delay, determine adjusting range:
;
Step 13: self adaptation is adjusted protocol parameter:
Step 14: modifying agreements parameter area:
4, the monolateral acceleration FAST TCP based on historical connection information according to claim 1 improves algorithm, it is characterized in that, when new FAST TCP is difficult to accurately obtain fast propagation delay, comprises the steps:
Step 1: the yardstick cycle is read the active current time queuing delay that connects between each hour
, settling time
And running time.
Step 2: the FAST of selection foundation the earliest connects from active connection
Queuing delay as bottleneck link
Step 3: when having newly-built connection to set up, provide the current queuing delay that connects with the active FAST TCP that sets up the earliest in history of the identical destination host of this newly-built connection
Step 4: to the FAST TCP of lower floor algorithm, when having newly-built FAST TCP connection to set up, revise the method that it calculates propagation delay for the first time:
Each new FAST TCP is connected:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012103994043A CN102904829A (en) | 2012-10-19 | 2012-10-19 | Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012103994043A CN102904829A (en) | 2012-10-19 | 2012-10-19 | Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102904829A true CN102904829A (en) | 2013-01-30 |
Family
ID=47576872
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012103994043A Pending CN102904829A (en) | 2012-10-19 | 2012-10-19 | Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102904829A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103246768A (en) * | 2013-04-28 | 2013-08-14 | 广东石油化工学院 | Modified overall stability analyzing method for solution trajectory bounds |
CN111586088A (en) * | 2020-03-11 | 2020-08-25 | 安超云软件有限公司 | Synchronous flow control method, system, device and readable storage medium |
CN114051057A (en) * | 2021-11-01 | 2022-02-15 | 北京百度网讯科技有限公司 | Method and device for determining queuing time of cloud equipment, electronic equipment and medium |
CN117692396A (en) * | 2024-02-04 | 2024-03-12 | 湖南国科亿存信息科技有限公司 | TCP unilateral acceleration method and device in complex network environment |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101023455A (en) * | 2004-08-17 | 2007-08-22 | 加州理工大学 | Method and apparatus for network congestion control using queue control and one-way delay measurements |
CN101056259A (en) * | 2007-05-21 | 2007-10-17 | 中南大学 | A collaborative congestion control method for high-speed network |
-
2012
- 2012-10-19 CN CN2012103994043A patent/CN102904829A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101023455A (en) * | 2004-08-17 | 2007-08-22 | 加州理工大学 | Method and apparatus for network congestion control using queue control and one-way delay measurements |
CN101056259A (en) * | 2007-05-21 | 2007-10-17 | 中南大学 | A collaborative congestion control method for high-speed network |
Non-Patent Citations (3)
Title |
---|
文宏等: "FAST TCP协议研究及性能分析", 《计算机工程与应用》, no. 11, 11 April 2006 (2006-04-11), pages 124 - 127 * |
陈晓龙等: "一种协同改进FAST TCP公平性算发放", 《广东工业大学学报》, vol. 27, no. 4, 31 December 2010 (2010-12-31), pages 4 - 8 * |
陈晓龙等: "非线性时变时滞FAST TCP系统低保守全局稳定性", 《控制理论以应用》, vol. 29, no. 4, 30 April 2012 (2012-04-30), pages 474 - 484 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103246768A (en) * | 2013-04-28 | 2013-08-14 | 广东石油化工学院 | Modified overall stability analyzing method for solution trajectory bounds |
CN111586088A (en) * | 2020-03-11 | 2020-08-25 | 安超云软件有限公司 | Synchronous flow control method, system, device and readable storage medium |
CN111586088B (en) * | 2020-03-11 | 2023-04-07 | 安超云软件有限公司 | Synchronous flow control method, system, device and readable storage medium |
CN114051057A (en) * | 2021-11-01 | 2022-02-15 | 北京百度网讯科技有限公司 | Method and device for determining queuing time of cloud equipment, electronic equipment and medium |
CN114051057B (en) * | 2021-11-01 | 2023-11-03 | 北京百度网讯科技有限公司 | Cloud equipment queuing time determination method and device, electronic equipment and medium |
CN117692396A (en) * | 2024-02-04 | 2024-03-12 | 湖南国科亿存信息科技有限公司 | TCP unilateral acceleration method and device in complex network environment |
CN117692396B (en) * | 2024-02-04 | 2024-04-26 | 湖南国科亿存信息科技有限公司 | TCP unilateral acceleration method and device in complex network environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106533960B (en) | A kind of data center network method for routing based on Fat-Tree structure | |
CN103346922B (en) | The controller of determination network state based on SDN and determine method | |
CN105471764B (en) | A kind of method of end-to-end QoS guarantee in SDN network | |
CN106160953B (en) | A kind of transmission method based on learning-oriented energy efficiency model | |
CN102075444B (en) | Network system and method for guaranteeing multi-type service quality | |
CN105897575A (en) | Path computing method based on multi-constrained path computing strategy under SDN | |
CN105515880B (en) | A kind of token bucket flow shaping method of suitable converged network | |
CN104767694A (en) | Data stream forwarding method facing Fat-Tree data center network architecture | |
CN111245722B (en) | SDN data center network flow forwarding method based on genetic algorithm | |
CN106302228B (en) | The transfer control method of task based access control perception in a kind of data center network | |
CN110191065A (en) | High-performance supported balanced System and method for based on software defined network | |
CN104993941A (en) | Openflow-based network highly-fault-tolerant virtual network mapping algorithm | |
CN106059951A (en) | Transmission control method for DCN (Data Center Network) based on multilevel congestion feedback | |
CN104092625B (en) | A kind of self adaptation being used in DCN asks dispatching method in batches | |
CN105490962A (en) | QoS management method based on OpenFlow network | |
CN114567598A (en) | Load balancing method and device based on deep learning and cross-domain cooperation | |
CN113271256B (en) | Information age multi-path transmission method and system | |
CN102904829A (en) | Unilateral acceleration FAST TCP (Fast Active queue management Scalable Transmission Control Protocol) improved algorithm based on history linkage information | |
CN105915467A (en) | Data center network flow balancing method and device oriented to software definition | |
CN102427596B (en) | Routing method and scheduling method of node mobile network assisted by positioning information | |
CN102356602A (en) | System and method of monitoring packets in flight for optimizing packet traffic in a network | |
CN102932840A (en) | Wireless router active queue management method based on rates and queue length | |
Zhang et al. | Adaptive marking threshold method for delay-sensitive TCP in data center network | |
CN107046507B (en) | A kind of jamming control method for multiplexed transport in DCN | |
CN104869113A (en) | Network transmission method and system based on minimum completion time |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130130 |