JP2008097312A - Multiplexer and information processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that it is difficult to fairly multiplex requests from a plurality of request sources. <P>SOLUTION: A storage multiplexer 30 includes a plurality of reservation channels 32 to which the request of each request source is input and an output port 42. The output port 42 outputs the request to be transferred from the plurality of reservation channels 32 to a disk scheduler 20. Each reservation channel 32 accepts the reservation of bandwidth for processing the request from each request source. Time frames as time units for multiplexing the requests from the plurality of request sources are prepared, and the reservation channel 32 whose reservation bandwidth has been at least partially seized in response to the request to be output from the other reservation channel 32 in a certain time frame receives the compensation of the seized reservation bandwidth in the next time frame. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a technique for multiplexing requests from a plurality of request sources.

  The operating system must support a wide range of applications, including real-time applications that use continuous media such as moving images and audio, file transfers from WWW (World Wide Web) servers and database servers, and interactive applications such as word processors. . These applications involve access to storage devices such as hard disks, but the quality of service (QoS) requirements are different. Interactive applications such as text editing require that the average response time of requests is minimal. On the other hand, file transfer requires maximum throughput. Also, multimedia applications such as video and audio require that frames of moving images arrive before the deadline, but to some extent allow frame dropping.

  In order to meet such a wide range of service quality requirements, it is necessary to appropriately schedule requests for individual applications in the disk device. A disk is generally assigned a logical block address by a combination of cylinders, tracks and sectors. The disk access time is determined by the seek time required for the disk head to move to the track position to be accessed, the rotation waiting time until the sector to be accessed reaches the disk head position, and the data read from the sector to be accessed. It is the sum of the data transfer time taken to transfer. A particularly dominant disk access time is a seek time, that is, a time required for moving the disk head, and how to reduce this determines the disk performance.

  Typical disk scheduling methods include SSTF (Shortest Seek Time First) algorithm and elevator algorithm. In the SSTF algorithm, processing is performed in order of seek time, that is, in order from the request with the smallest arm movement, and the movement distance of the disk head can be reduced. In the elevator algorithm, the request processing order is changed so that the arm moves in one direction, and the disk head is continuously moved from the inner circumference to the outer circumference or from the outer circumference to the inner circumference to reach the end. Do not change the moving direction of the disk head until this is done.

  The operating system needs to meet various service quality requirements for a wide range of applications, and the quality of data transfer from the disk device must meet the service quality requirements of the individual applications. However, scheduling requests with the highest priority on the quality of service of each application increases disk access time such as seek time and rotation latency, reduces disk performance, and reduces data transfer bandwidth from disk devices. It will be. On the other hand, if a scheduling method that maximizes the disk performance is adopted, requests for individual applications are scheduled in a manner that ignores the service quality requirements, so that the service quality requirements of the applications cannot be satisfied.

  When scheduling and executing requests from a plurality of request sources having different service quality requirements, it is necessary to multiplex requests from the plurality of request sources. If the request multiplexing process is not performed efficiently, the execution of the request will be hindered. Further, in multiplexing requests, it is necessary to ensure fairness among a plurality of request sources.

  The present invention has been made in view of such problems, and an object of the present invention is to provide a multiplexing processing technique capable of fairly multiplexing requests from a plurality of request sources.

  In order to solve the above problems, a multiplexer according to an aspect of the present invention includes a plurality of channels and an output port. The channel is provided for each request source of the service, accepts a bandwidth reservation for processing the request from each request source, and receives a request from each request source. The output port outputs requests passed from a plurality of the channels. A channel in which a time frame that is a time unit for multiplexing requests from multiple request sources is provided, and at least a part of the reserved bandwidth is intercepted by a request output from another channel in a certain time frame Will be compensated for the intercepted reserved bandwidth in the next time frame.

  Another aspect of the present invention is an information processing apparatus. The apparatus includes a multiplexer that outputs a request input from one or more request sources of a service, and a scheduler that schedules and executes the request output from the multiplexer. The multiplexer includes a plurality of channels and an output port. The channel is provided for each request source, receives a bandwidth reservation for processing a request from each request source, and receives a request from each request source. The output port outputs requests passed from a plurality of the channels. A channel in which a time frame that is a time unit for multiplexing requests from multiple request sources is provided, and at least a part of the reserved bandwidth is intercepted by a request output from another channel in a certain time frame Will be compensated for the intercepted reserved bandwidth in the next time frame.

  Here, the “request” is not limited to a request for a storage device such as a hard disk but includes a request for various devices such as a processor, a memory, and a peripheral device, and a request for a device such as a server via a network. In addition to disk scheduling, “scheduling” includes access scheduling for various devices such as processors, memories, and peripheral devices, and access scheduling for devices such as servers via a network.

  It should be noted that any combination of the above-described constituent elements and a representation of the present invention converted between a method, an apparatus, a system, a computer program, a data structure, etc. are also effective as an aspect of the present invention.

  According to the present invention, it is possible to improve the quality of service by fairly multiplexing requests from a plurality of request sources.

[1] Configuration FIG. 1 is a diagram illustrating the configuration of the storage subsystem 100. The storage subsystem 100 is provided with a file system 130 in the operating system 140.

  The file system 130 establishes a connection with the storage subsystem 100 via the storage interface 120, and supplies an access request for the storage device 150 to the storage subsystem 100 using the established connection. The file system 130 can establish a plurality of connections with the storage subsystem 100. The file system 130 may set a connection for each application type. Application types include multimedia applications, real-time applications, interactive applications, and file transfers.

  The storage subsystem 100 includes a storage service provider 60, a SCSI (Small Computer System Interface) class driver 50, a request scheduler 40, and a device driver 10. The storage service provider 60 manages a connection from the file system 130 and provides a storage access service to the file system 130. The storage service provider 60 can reserve the data transfer bandwidth of the storage device 150 for each connection according to the application type or a request from the application.

  The SCSI class driver 50 is a driver according to the SCSI standard, issues an access request to the storage device 150 from each connection, and supplies it to the request scheduler 40.

  The request scheduler 40 includes a storage multiplexer 30 and a disk scheduler 20, and multiplexes requests from one or more connections and gives them to the device driver 10.

  The device driver 10 is a device-specific driver having device information of the storage device 150. The device driver 10 executes a request passed from the disk scheduler 20, reads / writes data from / to the storage device 150, and notifies the SCSI class driver 50 of the result.

  FIG. 2 is a configuration diagram of the request scheduler 40. The request scheduler 40 multiplexes and schedules requests for the storage device 150. This figure depicts a block diagram focusing on functions, and these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof.

  The request scheduler 40 includes a storage multiplexer (hereinafter simply referred to as “multiplexer”) 30 and a disk scheduler 20. The multiplexer 30 multiplexes requests from each connection in “time-frame” units described later, and secures the bandwidth of each connection. The disk scheduler 20 schedules the execution order of the requests supplied from the multiplexer 30 in units of time frames and supplies them to the device driver 10.

  The multiplexer 30 includes a plurality of channels (reference numerals 31 and 32) into which requests for each connection are input, an output port 42, a timer 38, and a free bandwidth port 44.

  The channel is a logical path provided in the multiplexer 30 through which requests are input and output, and controls the output of requests input to the multiplexer 30 by opening and closing ports in the channel. The bandwidth of a channel is given by the maximum data transfer rate from that channel. There are two types of channels: main channel 31 and reserved channel 32.

  The main channel 31 is a connection channel that does not reserve a bandwidth. For the data transfer from the main channel 31, the remaining unreserved bandwidth out of the total bandwidth that can be reserved (“total reserved bandwidth net_bw” described later) is used. If no reserved channel 32 is generated, the main channel 31 can use the entire band for data transfer.

  The reservation channel 32 is provided for each connection that reserves bandwidth. The reservation channel 32 is generated at runtime each time a connection for reserving bandwidth is established, and is discarded when the connection disappears.

  The main channel 31 has a channel port 33 and a channel queue 35. Similarly, the reserved channel 32 has a channel port 34 and a channel queue 36. A request for each connection is input from the SCSI class driver 50 to the channel ports 33 and 34 of the channels 31 and 32. When the channel ports 33 and 34 are open, the request input to the channel ports 33 and 34 is sent directly to the output port 42 without being queued to the channel queues 35 and 36. Channel ports 33 and 34 are closed after a reserved amount of requests has passed through the ports. If the channel ports 33, 34 are closed, the request is queued in the channel queues 35, 36.

  Each channel queue 35, 36 accumulates requests input to the channel ports 33, 34 while the channel ports 33, 34 are closed for the current time frame. The channel port 34 of the reserved channel 32 opens at the start of each time frame and closes when the reserved amount of requests passes through the port, so that only the reserved amount of requests is sent to the output port 42. The channel port 33 of the main channel 31 is also opened at the start of each time frame, and is closed when the amount of requests that can be accommodated within the unreserved bandwidth out of the total reserved bandwidth passes through the port. Only requests in the bandwidth range are sent to the output port 42.

  The timer 38 measures the passage of time for one time frame using a clock, and notifies the main channel 31, the reserved channel 32, and the output port 42 of the passage of one time frame. In response to the notification from the timer 38, various parameters are appropriately reset at the start of a new time frame in the main channel 31, the reserved channel 32, and the output port 42. When the timer 38 notifies that one time frame has elapsed, the channel ports 33 and 34 are opened in the channels 31 and 32, and first, the requests queued in the channel queues 35 and 36 of the channels 31 and 32 are opened. Sent from the channel ports 33 and 34 to the output port 42.

  The output port 42 multiplexes the requests passed from the channels 31 and 32 and sends them to the disk scheduler 20. As a result, the request input to the multiplexer 30 is multiplexed through the output port 42 and passed to the disk scheduling queue 22 (hereinafter referred to as “disk scheduler queue”) 22 of the disk scheduler 20. It should be noted that a request that has been waiting in the channel queues 35 and 36 at the start of the time frame is output to the output port 42 from the opened channel ports 33 and 34, and the disk scheduler queue via the output port 42. 22 is sent. Requests are queued in the disk scheduler queue 22 in units of time frames. In the figure, a state in which requests 24a to 24c are queued in time frame n and requests 25a to 25d are queued in time frame n + 1 is schematically shown.

  The disk scheduler 20 receives a request multiplexed in units of frames from the multiplexer 30, queues the requests in the disk scheduler queue 22, and schedules the execution order of the requests in the frame for each frame. In order to maximize the disk performance, the disk scheduler 20 performs scheduling in consideration of the disk access pattern of each request in the frame, and supplies the scheduled request to the device driver 10.

  When there is no request in the disk scheduler queue 22, the disk scheduler 20 requests a request from the free bandwidth port 44. In response to the request from the disk scheduler 20, the multiplexer 30 selects the highest priority channel that queues the request, extracts the request queued in that channel, and passes it to the free bandwidth port 44. The disk scheduler 20 receives a request from the free bandwidth port 44 and executes it (that is, passes the request to the device driver 10).

As described above, the request input to the multiplexer 30 is passed to the disk scheduler 20 through one of the following three paths.
(1) When the channel ports 33 and 34 are open, the request is not directly queued in the channel queues 35 and 36 (bypassing the channel queues 35 and 36), but directly passed to the output port 42 and output. A path queued from the port 42 to the disk scheduler queue 22.
(2) When the channel ports 33 and 34 are closed, the request is once queued in the channel queues 35 and 36, and then from the opened channel ports 33 and 34 to the output port 42 at the beginning of the next time frame. A path that is passed and queued from the output port 42 to the disk scheduler queue 22.
(3) When the disk scheduler queue 22 is free, the requests queued in the channel queues 35 and 36 of the channel having a high priority are passed to the free bandwidth port 44 and are queued in the disk scheduler queue 22 Not the path to be executed.

  The internal configuration of the main channel 31 and the reserved channel 32 and the request input / output control will be described in detail with reference to FIGS. The internal configuration and input / output control of the main channel 31 and the reserved channel 32 are common and will be described together. As shown in FIG. 3A, the channel ports 33 and 34 of the channels 31 and 32 include an input side port 45 and an output side port 46. Each of the input side port 45 and the output side port 46 has an open terminal and a closed terminal, and the ports are opened and closed by switching to either one.

  Depending on the open / closed states of the input port 45 and the output port 46 of the channels 31 and 32, three channel port states as shown in FIG. State A is a state in which the input side port 45 is “open” and the output side port 46 is “closed”, and a request input to the open terminal of the input side port 45 passes through the closed terminal of the output side port 46. It is queued in the channel queues 35 and 36.

  State B is a state in which the input side port 45 is “open” and the output side port 46 is “open”, and a request input to the open terminal of the input side port 45 passes through the open terminal of the output side port 46. The data is output to the output port 42 (bypass the channel queues 35 and 36).

  In state C, the input side port 45 is “closed” and the output side port 46 is “open”, and the input side port 45 is closed. Therefore, no new request is input, and the channel queue 35 , 36 are input to the closed terminal of the input side port 45, and output to the output port 42 through the open terminal of the output side port 46.

  “Channel ports 33 and 34 are closed” means state A. At this time, the input request is queued in the channel queues 35 and 36.

  “Channel ports 33 and 34 are open” refers to two states, state B and state C. When a new time frame starts, the channel ports 33 and 34 are initially in the state C, and the requests queued in the channel queues 35 and 36 are output from the channel ports 33 and 34 to the output port 42. After this, if the channel bandwidth still remains, the channel ports 33, 34 will be in state B, and a newly arriving request will bypass the channel queues 35, 36 and pass from the channel ports 33, 34 to the output port 42. Is output. If the channel bandwidth is used up by outputting the queued requests in the state C from the channel ports 33 and 34, the channel ports 33 and 34 are in the state A, and the newly arriving request is the channel. It is queued in the queues 35 and 36.

  As described above, the channel ports 33 and 34 have the input side port 45 and the output side port 46 as internal configurations, and the input / output path of the request is controlled according to the open / closed state thereof. Therefore, it should be noted that the open / closed states of the channel ports 33 and 34 are described by focusing only on the open / closed state of the output port 46, without touching the control operation of the internal configuration and the request input / output path.

[2] Parameters Various parameters used in the request scheduler 40 of FIG. 2 will be described.
[2.1] Time Frame The time frame is a time unit in which the bandwidth is calculated in the multiplexer 30. The time frame is also a time window that divides requests multiplexed in the disk scheduler 20. The order is not exchanged between requests issued in different time frames, and the requests are executed in the order in which they are issued. The order may be changed in the disk scheduler 20 between requests issued within one time frame.

  The time frame is a variable value in the system. The longer the time frame, the better the throughput, and the shorter the time frame, the better the average response time. There is a trade-off between response time and throughput, and the time frame can be appropriately changed depending on the type of connection to be scheduled and the combination thereof.

[2.2] Reservation parameter The reservation parameter is a parameter used for bandwidth reservation, and is adjusted in accordance with the change of the bandwidth reservation. Note that the following bandwidth definition is scaled by the value of the time frame. The reservation parameters are as follows.

[2.2.1] Minimum Device Bandwidth and Maximum Device Bandwidth The data transfer rate that can be provided by the storage device 150 is referred to as device bandwidth. Since the seek time of the disk fluctuates, the storage device 150 cannot always perform the same amount of data transfer per unit time. The data transfer rate varies between a minimum value and a maximum value. The minimum device bandwidth is the minimum bandwidth that the storage device 150 can provide. Based on the worst case evaluation, the device driver 10 can determine and notify the minimum device bandwidth. The multiplexer 30 permits the bandwidth reservation in such a range that the total reserved bandwidth does not exceed the minimum device bandwidth. The maximum device bandwidth is a sustained data transfer rate that the storage device 150 can provide under the best conditions.

[2.2.2] Total Reservable Bandwidth, Total Reserved Bandwidth, and Total Unreserved Bandwidth Total reserved bandwidth net_bw is the total bandwidth that can be reserved by one or more channels, and is the minimum device bandwidth Equal to width. The total reserved bandwidth net_reserved is the total reserved bandwidth of all channels. The total unreserved bandwidth net_unreserved is a remaining bandwidth that can be reserved, and is equal to a value obtained by subtracting the total reserved bandwidth net_reserved from the total reserved bandwidth net_bw.

[2.3] Time Frame Parameter The time frame parameter is a parameter used in the multiplexer 30. The time frame parameter is a parameter common to all channels, and changes for each time frame. The time frame parameters are as follows.

[2.3.1] Temporary free bandwidth net_credit
The disk bandwidth that is not requested from any channel is called the provisional free bandwidth net_credit. A connection that reserves bandwidth does not necessarily use the bandwidth reserved in each time frame. Even a connection with reserved bandwidth may not request bandwidth within the time frame, that is, may not issue a request, resulting in free bandwidth. The difference between the minimum device bandwidth and the maximum device bandwidth is also added to the provisional free bandwidth net_credit for the current time frame.

  Requests for each connection may arrive at the beginning of the time frame or near the end of the time frame, and the arrival time of the request is unpredictable. Therefore, in the middle of a time frame, if a request from a certain reserved channel has not arrived and a bandwidth is vacant, it is assumed that a vacant bandwidth has been created for the time being. Allocate bandwidth for requests from. However, a request arrives late from the reserved channel within the time frame, and the reserved bandwidth may be required. In other words, because the reservation request does not arrive, assuming that it is free bandwidth, and using that free bandwidth for requests on other channels, there is not enough bandwidth allocated for the reservation channel requests that arrived late. It happens that the request cannot be executed. This is the state in which the “false free bandwidth” described below has occurred.

[2.3.2] False free bandwidth net_deficit
If one or more reserved channel requests become unexecutable due to lack of bandwidth in the current time frame, it means that it was an error to assume that there was "free bandwidth" in the current time frame. To do. The assumed free bandwidth was actually a “pseudo free bandwidth”, so the channel that requested the bandwidth must return the bandwidth to the deprived channel in the next time frame. I must. The bandwidth that has been used by assuming that it is an empty bandwidth by mistake is called a false empty bandwidth net_deficit.

[2.3.3] Total waste bandwidth net_wasteed
When neither multiplexer 30 nor disk scheduler 20 has a request, device driver 10 is idle and disk bandwidth is wasted. The bandwidth while the device driver 10 is idle, that is, the storage device 150 is idle is referred to as the total wasted bandwidth net_waste.

[2.3.4] Total bandwidth used net_used
The total used bandwidth net_used is the total size (in bytes) of requests that have passed through the output port 42 in the current time frame.

[2.3.5] Total unused bandwidth net_unused
The total unused bandwidth net_unused is the remaining amount of bandwidth that can be used in the current time frame, and is a value obtained by subtracting the total wasted bandwidth net_waste and the total used bandwidth net_used from the total reservable bandwidth net_bw.
net_unused = net_bw-(net_wasted + net_used)

[2.4] Channel Parameter The channel parameter is also a parameter used in the multiplexer 30. The channel parameter is a parameter set for each channel and indicates a channel state. Channel parameters include the following:

[2.4.1] Channel reserved bandwidth c_reserved
The bandwidth allocated to the channel is referred to as channel reserved bandwidth c_reserved.
[2.4.2] Channel usage bandwidth c_used
The channel use bandwidth c_used is the total size (in bytes) of requests that have passed through the channel in the current time frame.
[2.4.3] Channel unused bandwidth c_unused
The remaining amount of bandwidth that the channel can use for data transfer in the current time frame. The unused channel bandwidth c_unused is the reserved channel bandwidth c_reserved minus the used channel bandwidth c_used and the channel wasted bandwidth c_waste described below.
c_unused = c_reserved-(c_used + c_wasted)

[2.4.4] Channel wasted bandwidth c_wasteed
Of the total wasted bandwidth net_waste, the portion allocated to each channel is referred to as channel wasted bandwidth c_wasteed. The channel waste bandwidth c_wasteed of each channel is proportional to the channel unused bandwidth c_unused of that channel. That is, the total wasted bandwidth net_waste is distributed among the channels at a ratio of the unused channel bandwidth c_unused.

[2.4.5] Channel credit bandwidth c_credit
If the channel uses more than the allocated bandwidth, the channel credit bandwidth c_credit is incremented by the excess amount. When a channel is deprived of the bandwidth allocated to it by another channel, the channel credit bandwidth c_credit is decremented by the deprived amount. A positive channel credit bandwidth c_credit value means that the channel has used more than the reserved amount, and a negative channel credit bandwidth c_credit value uses less than the reserved amount for the channel. Means that.

  The channel credit bandwidth c_credit is used to set the channel usage bandwidth c_used for the next time frame. If a channel steals the bandwidth of another channel and uses more bandwidth, in the next time frame, that channel will use less bandwidth to make up for the excess. If a channel is deprived of the bandwidth that it uses, it will be compensated for the channel credit bandwidth c_credit in the next time frame.

[2.4.6] Channel priority c_priority
The channel priority c_priority is a priority when a channel uses an empty bandwidth. The higher the priority, the higher the probability that the channel can acquire free bandwidth. A method for adjusting the channel priority c_priority will be described later.

[2.5] Bandwidth Reservation Type Each connection reserves the absolute bandwidth of the data transfer bandwidth of the storage device 150. The sum of the bandwidths reserved for all connections cannot exceed the total reservable bandwidth net_bw of the storage device 150. By reserving the bandwidth, it is guaranteed that a certain amount of data is transferred at a minimum within a certain unit time, so that the response time can be prevented from becoming unpredictable. In addition, by reserving the bandwidth, the data transfer capability of the storage device 150 can be shared fairly between the connections.

  There are two types of bandwidth reservations: reservations that specify a minimum bandwidth and reservations that specify a maximum bandwidth. In the minimum bandwidth reservation, the specified amount of bandwidth is treated as a minimum request. A connection that reserves a minimum bandwidth is guaranteed a specified amount of bandwidth, and a higher data transfer rate may be obtained as long as it does not affect other bandwidth reservations.

  In maximum bandwidth reservation, a specified amount of bandwidth is treated as a maximum request. A connection that reserves the maximum bandwidth will never have a data transfer rate that exceeds the specified amount. Maximum bandwidth reservation is used to limit a channel from exceeding the maximum bandwidth and requesting free bandwidth.

[3] Request Multiplexing Process [3.1] Overview A request multiplexing process procedure by the request scheduler 40 will be described. The request multiplexing process includes a process for compensating for an error in bandwidth allocation (the aforementioned “false free bandwidth”). Bandwidth allocation errors that occur in one time frame are compensated for in the next time frame. By using the error compensation process, the multiplexing process by the multiplexer 30 can be made light in weight while maintaining high accuracy. Since the desired multiplexing processing behavior is realized across two frames, the unit time for which complete multiplexing processing is guaranteed is a length of two frames.

The request multiplexing process is based on the following three policies.
(1) A request input to a channel port is allowed to pass through the channel port until the channel reserved bandwidth c_reserved is reached. After reaching the channel reserved bandwidth c_reserved, a request input to the channel port is queued in the channel queue. Queued requests are limited to using bandwidth in the current time frame.
(2) When the current time frame ends and the next time frame starts, a request is output from the channel queue (with the channel unused bandwidth c_unused as the upper limit).
(3) When the disk scheduler queue 22 is empty, the disk scheduler 20 can retrieve requests one by one from the channel with the highest priority of the multiplexer 30.

[3.2] Channel Port Status A channel port has either an open state or a closed state depending on the value of the channel unused bandwidth c_unused. If the value of channel unused bandwidth c_unused is positive (not including zero), the channel port is open, otherwise it is closed. When the channel port is open, a request input to the channel port is sent directly to the output port 42. When a predetermined amount of requests pass through the channel port, the channel port is closed until the next time frame. When the channel port is closed, the request is queued in the channel queue. A channel port is used to restrict that channel from using the bandwidth of other channels.

[3.3] Status of Output Port The output port 42 is in an open state or a closed state depending on the value of the unused bandwidth net_unused. If the value of the unused bandwidth net_unused is positive, the output port 42 is open, otherwise it is closed.

[3.4] Adjustment of Channel Priority When the available bandwidth is available, the multiplexer 30 assigns the available bandwidth to the channel with the highest priority. The channel priority c_priority is the sum of the channel basic priority c_base_priority and the number of channel queue requests c_queued_request.
c_priority = c_base_priority + c_queued_request

The channel basic priority c_base_priority distinguishes channels according to the difference in channel type or the degree of urgency, and has the following three levels.
Normal level: Default priority for each channel.
Priority level: It is set for a channel for which it is preferable to use a free bandwidth.
Deprivation level: When a channel cannot use its reserved bandwidth in the current time frame, the priority of the deprivation level is set by the multiplexer 30 in the next time frame.
The priority increases in the order of normal level, priority level, and deprivation level. Multiplexer 30 resets the channel basic priority when a new time frame begins.

  The number of channel queue requests c_requested_request is the number of requests pending in the channel queue. As the number of requests in the channel queue increases, the channel priority increases.

[3.5] Wasted bandwidth update If the disk scheduler 20 is idle, bandwidth is wasted until a request arrives. The multiplexer 30 records the time stamp waste_time_start when the disk scheduler 20 becomes idle.

When a request arrives when the disk scheduler 20 is idle, the multiplexer 30 obtains the current timestamp waste_time_end and calculates the wasted time wasted_time.
waste_time_end = get_current_time ();
wasted_time = waste_time_end-waste_time_start;

The multiplexer 30 calculates the wasted bandwidth wasted_bandwidth corresponding to the wasted time using the minimum device bandwidth, and updates the total wasted bandwidth net_wasteed.
wasted_bandwidth = scale_to_bandwidth (wasted_time);
net_wasted = net_wasted + wasted_bandwidth;

The total wasted bandwidth net_waste is distributed among all channels with non-zero channel unused bandwidth c_unused in proportion to the channel unused bandwidth c_unused of each channel. That is, the channel wasted bandwidth c_wasteed is calculated as follows.
for all channels:
c_wasted = c_wasted + (c_unused * wasted_bandwidth) / net_unused;

  In this way, the multiplexer 30 adds the channel waste bandwidth c_waste obtained by proportionally allocating the total waste bandwidth net_waste to the channel use bandwidth c_used of each channel for all channels scheduled to issue requests. A penalty is given to each channel by treating the channel as if it had used the channel wasted bandwidth c_waste. The channel unused bandwidth c_unused is updated and decreased by the updated channel wasted bandwidth c_waste. Based on the updated channel unused bandwidth c_unused, the multiplexer 30 updates the state of the channel port. When the channel unused bandwidth c_unused becomes 0 or less, the channel port is closed.

[3.6] Flowchart A procedure of request multiplexing processing by the request scheduler 40 will be described with reference to FIGS. FIG. 4 is a flowchart showing the overall flow of the request multiplexing process. The request scheduler 40 is in an event generation waiting state (S100). There are three events, event A “request arrival”, event B “disk scheduler queue is empty”, and event C “start of new time frame”. When each event occurs, processing in steps S110, S120, and S130 is performed. Are executed respectively. When each process is executed, the request scheduler 40 confirms the stop condition (S140). If the request scheduler 40 does not stop (N in S140), the process returns to the event occurrence waiting state in step S100 and stops (Y in S140). To do.

[3.6.1] Event A: Request Arrival FIG. 5A is a flowchart showing processing when event A “request arrival” in step S110 occurs. If the disk scheduler queue 22 was empty before the arrival of a new request (Y in S9), the current time is recorded to end the measurement of the wasted bandwidth (S10), and the wasted bandwidth is updated. The channel port status is updated (S11). If the disk scheduler queue 22 has not been empty before the arrival of a new request (N in S9), the waste bandwidth is not measured, so the processing in steps S10 and S11 is unnecessary. Proceed to S12.

  FIG. 5B is a flowchart showing in detail the processing of updating the wasted bandwidth and updating the state of the channel port in step S11.

The total wasted bandwidth net_waste and the channel wasted bandwidth c_waste are updated according to the procedure described in [3.5] above (S150). The channel unused bandwidth c_unused is updated by the following equation based on the value of the channel used bandwidth c_used and the updated value of the channel wasted bandwidth c_used (S152). If the channel unused bandwidth c_unused is not positive (S154) N) The channel port is closed (S156).
c_unused = c_reserved-(c_used + c_wasted)

The total unused bandwidth net_unused is updated by the following equation based on the value of the total used bandwidth net_used and the updated value of the total wasted bandwidth net_used (S158), and if the total unused bandwidth net_unused is not positive (S160 N), the output port 42 is closed (S162).
net_unused = net_bw-(net_used + net_wasted)

  Returning to FIG. 5A. If the channel port of the channel on which the request has arrived is closed (N in S12), the request is queued in the channel queue, the channel priority is incremented by one (S38), and the processing of event A ends.

  If the channel port of the channel on which the request has arrived is open (Y in S12) and the output port 42 is open (Y in S14), the request is sent directly to the output port 42 and output from the output port 42. (S16).

When the request is output from the output port 42, the channel use bandwidth c_used and the total use bandwidth net_used are incremented by the request size (byte unit) according to the following equation (S18).
c_used = c_used + request_size
net_used = net_used + request_size

The channel unused bandwidth c_unused is updated by the following equation with the updated value of the channel used bandwidth c_used (S20). If the channel unused bandwidth c_unused is not positive (N in S22), the channel port is closed. (S24).
c_unused = c_reserved-(c_used + c_wasted)

The total unused bandwidth net_unused is updated by the following equation with the updated value of the total used bandwidth net_used (S26). If the total unused bandwidth net_unused is not positive (N in S28), the output port 42 is closed. (S30).
net_unused = net_bw-(net_used + net_wasted)

  In step S18, when the request size is larger than the channel unused bandwidth c_unused, it is necessary to allocate a bandwidth exceeding the channel unused bandwidth c_unused, and therefore, the channel credit bandwidth is increased by the excess bandwidth. After incrementing the width c_credit, the channel use bandwidth c_used and the total use bandwidth net_used are incremented by the request size (in bytes). In this case, the channel port is immediately closed. If step S18 is written in the pseudo code, it becomes as follows.

if (c_unused <request_size)
c_credit = c_credit + (request_size-c_unused);
c_used = c_used + request_size;
net_used = net_used + request_size;

  In step S14, when the output port 42 is closed (N in S14), it means that the output port 42 is closed even though the channel port of the channel on which the request arrived is open. It means that the channel used “fake free bandwidth”. In this case, the request is queued in the channel queue, and the basic priority of the channel is set to “stripping level” (S32).

  Next, the deprived bandwidth deprived is obtained (S34). The deprived bandwidth deprived is usually the request size, but the request size may exceed the channel unused bandwidth c_unused, so the deprived bandwidth deprived is exactly the request size and the channel unused It is given by the smaller of the bandwidth c_unused. The deprived bandwidth deprived is subtracted from the channel credit bandwidth c_credit, and the deprived bandwidth deprived is added to the false free bandwidth net_defict (S36). The pseudo code in steps S34 and S36 is as follows.

if (c_unused> 0) {
deprived = smallerof (c_unused, request_size);
c_credit = c_credit-deprived;
net_deficit = net_deficit + deprived;
}

[3.6.2] Event B: Disk Scheduler Queue is Empty FIG. 6 is a flowchart showing processing when event B “Disk Scheduler Queue is empty” in step S120 occurs. When the multiplexer 30 outputs a request that is less than the actual device bandwidth, the disk scheduler queue 22 becomes free in each time frame. The disk scheduler queue 22 becomes empty even when there is no input request. Event B “disk scheduler queue is empty” is an event that occurs when the disk scheduler 20 attempts to retrieve a request from the free bandwidth port 44 when the disk scheduler queue 22 is empty.

Upon receiving access to the free bandwidth port 44 from the disk scheduler 20, the multiplexer 30 checks whether there is a request in the channel queue of any channel (S50). If there is a request in any of the channel queues, the request is taken out from the channel queue with the highest priority and passed to the free bandwidth port 44 (S52). The multiplexer 30 increments the channel credit bandwidth c_cred and the provisional free bandwidth net_credit by the request size (S54).
c_credit = c_credit + request_size
net_credit = net_credit + request_size

  The output port 42 is notified that a request has been acquired from the free bandwidth port 44, the total used bandwidth net_used is updated, and as a result, the total unused bandwidth net_unused is updated (S55). If the updated total unused bandwidth net_unused is not positive (N in S56), the output port 42 is closed (S57).

  The disk scheduler 20 acquires a request from the free bandwidth port 44 and proceeds to execution (S58).

  Whether the bandwidth is really free is unknown until the time frame ends and a new time frame begins, so steps S52-S58 are performed under the assumption of "free bandwidth". If it is “false free bandwidth”, the error is compensated for by using the channel credit bandwidth in the next time frame. The error compensation process will be described in detail with reference to FIGS.

  If there is no request in any channel queue (N in S 50), the multiplexer 30 does not pass the request to the free bandwidth port 44. At this time, the multiplexer 30 can know that the disk scheduler 20 is idle because the disk scheduler 20 was trying to fetch a request from the free bandwidth port 44. Therefore, the multiplexer 30 records the current time to start measuring the wasted bandwidth (S59). As already described in FIG. 5, when the next request arrives in the current time frame, the multiplexer 30 finishes measuring the wasted bandwidth, records the time at that time, and sets the channel for all channels. The wasted bandwidth c_used is updated, and the state of the channel port is appropriately changed before processing a new request (see S9 to S11 in FIG. 5A).

[3.6.3] Event C: Start of New Time Frame FIG. 7 is a flowchart showing processing when event C “Start of new time frame” in step S130 occurs. When the period of the previous time frame TF (n-1) ends, a new time frame TF (n) starts (where n indicates the number of the time frame). When a new time frame TF (n) starts, the multiplexer 30 first checks whether the pseudo free bandwidth net_deficit is zero (S60). If the false free bandwidth net_defict is zero (Y in S60), the multiplexer 30 resets the channel use bandwidth c_used and the channel credit bandwidth c_credit of all the channels to zero, and uses the total use bandwidth net_used and the provisional free bandwidth net_credit. Is also reset to zero (S62), and the process proceeds to step S65.

  If the false free bandwidth net_defict is not zero (N in S60), the channel using the temporary free bandwidth net_cred in the previous time frame TF (n-1) uses the bandwidth in the current time frame TF (n). The amount must be reduced. By increasing the channel use bandwidth c_used for the channel that used the excess bandwidth in the previous time frame TF (n−1), error compensation for the pseudo free bandwidth net_defit is performed (S64). The error compensation process in step S64 will be described in detail with reference to FIG.

  When the error compensation process is completed, the output port 42 is opened (S65). The channel unused bandwidth of each channel is checked (S66). If the channel unused bandwidth is positive (Y in S66), the channel port of that channel is opened (S68), and as long as the channel port continues to be opened, Until the channel queue is emptied, the request waiting in the channel queue is immediately transferred from the opened channel port to the output port 42 and output from the output port 42 (S70). Each time a request is output from the channel queue, the unused channel bandwidth of that channel decreases. In step S66, if the channel unused bandwidth of the channel is not positive (N in S66), the channel port is closed and is not opened in a new time frame TF (n) (S72).

  FIG. 8 is a flowchart showing a detailed procedure of the error compensation process in step S64. If the lowest priority channel is selected (S80) and the false free bandwidth net_deficit is positive (Y in S82), steps S84 to S92 are executed for the selected channel, and the false free bandwidth net_deficit becomes 0. If (N in S82), step S94 is executed for the selected channel. If there is a channel with the next lowest priority (Y in S96), the channel is selected (S96: Y). Thereafter, steps S82 to S94 are repeated. If all channels have been processed (N in S96), the error compensation process is terminated.

  First, the processing of steps S84 to S92 for the selected channel will be described for the case where the false free bandwidth net_deficit is positive (Y in S82).

  For the selected channel, if the channel credit bandwidth c_credit is positive (Y in S84) and the channel credit bandwidth c_credit is smaller than the fake free bandwidth net_deficit (Y in S86), the fake free bandwidth net_deficit is set. The channel credit bandwidth c_cred is subtracted (S88). Then, the channel credit bandwidth c_cred is substituted for the channel use bandwidth c_used, and the channel credit bandwidth c_cred is set to zero (S92). As a result, this channel is in a state where the bandwidth is used as much as the channel credit bandwidth c_credit.

  If the channel credit bandwidth c_cred is positive (Y in S84) but the channel credit bandwidth c_cred is equal to or larger than the false free bandwidth net_deficit (N in S86), the channel credit bandwidth c_cred is replaced with the false free bandwidth net_deficit. The false free bandwidth net_deficit is set to zero (S90). After that, the channel credit bandwidth c_cred is substituted for the channel use bandwidth c_used to set the channel credit bandwidth c_cred to zero (S92). Thereby, in the case of the channel credit bandwidth c_credit exceeding the fake free bandwidth net_defit, the channel credit bandwidth c_credit is once reduced to the value of the fake free bandwidth net_deficit and then substituted into the channel use bandwidth c_used.

  If the channel credit bandwidth c_cred is not positive (N in S84), the channel credit bandwidth c_cred is substituted for the channel usage bandwidth c_used, and the channel credit bandwidth c_cred is set to zero (S92). Therefore, when the channel credit bandwidth c_cred is negative, the channel use bandwidth c_used is negative, and the channel unused bandwidth c_unsed is increased by that amount, so that it is possible to use extra bandwidth.

  When the false free bandwidth net_deficit becomes zero (N in S82), no further error compensation processing is required, so that the channel credit bandwidth c_credit and the channel use bandwidth c_used are set to zero for the selected channel (S94). ).

  FIG. 9 is a diagram illustrating the procedure of the error compensation process of FIG. 8 using a specific example. The figure shows how the channel usage bandwidth c_used of each channel of the current time frame is determined based on the channel credit bandwidth c_cred of each channel in the previous time frame.

  The channel credit bandwidths c_credit of the six channels C1 to C6 in the previous time frame are 2, 4, 3, 1, -4, and -1, respectively. The temporary free bandwidth net_credit in the previous time frame is the sum of the positive channel credit bandwidths 2, 4, 3, 1 of the channels C1 to C4, and its value is 10. On the other hand, the pseudo free bandwidth net_deficit in the previous time frame is the sum of the absolute values of the negative channel credit bandwidths −4 and −1 of the channels C5 and C6, and the value thereof is 5. Therefore, the bandwidth that must be compensated in the current time frame is 5.

  Assume that the channels C1 to C6 have lower priority in this order. Since the lowest priority channel C6 is selected (S80), the false free bandwidth net_deficit is 5 (Y in S82), and the channel credit bandwidth c_credit of the channel C6 is -1 (N in S84), so that the channel C6 The channel usage bandwidth c_used becomes −1 by substituting the value of the channel credit bandwidth c_credit (S92).

  Next, the channel C5 with the lowest priority is selected (S96), the false free bandwidth net_deficit is 5 (Y in S82), and the channel credit bandwidth c_credit of the channel C5 is −4 (N in S84). The channel use bandwidth c_used of the channel C5 becomes −4 when the value of the channel credit bandwidth c_credit is substituted (S92).

  Further, the channel C4 having the next lowest priority is selected (S96), the false free bandwidth net_deficit is 5 (Y in S82), the channel credit bandwidth c_credit of the channel C4 is 1 (Y in S84), and the channel Since the value 1 of the credit bandwidth c_credit is smaller than the value 5 of the false free bandwidth net_defict (Y of S86), the false free bandwidth net_defit is subtracted by the value 1 of the channel credit bandwidth c_credit to 4 (S88). Then, the channel use bandwidth c_used of the channel C4 becomes 1 when the value of the channel credit bandwidth c_credit is substituted (S92).

  Further, the channel C3 having the next lowest priority is selected (S96), the false free bandwidth net_deficit is 4 (Y in S82), the channel credit bandwidth c_credit of the channel C3 is 3 (Y in S84), and the channel Since the value 3 of the credit bandwidth c_credit is smaller than the value 4 of the false free bandwidth net_defict (Y of S86), the false free bandwidth net_defit is subtracted by 1 by the value 3 of the channel credit bandwidth c_credit. (S88). Then, the channel use bandwidth c_used of the channel C3 becomes 3 when the value of the channel credit bandwidth c_credit is substituted (S92).

  Further, the channel C2 having the next lowest priority is selected (S96), the false free bandwidth net_deficit is 1 (Y in S82), the channel credit bandwidth c_credit of the channel C2 is 4 (Y in S84), and the channel Since the value 4 of the credit bandwidth c_credit is not smaller than the value 1 of the fake free bandwidth net_defict (N in S86), the channel credit bandwidth c_credit is replaced with the value 1 of the fake free bandwidth net_deficit, and the fake free bandwidth net_deficit Is set to zero (S90). After that, the channel use bandwidth c_used of the channel C2 becomes 1 by substituting the value 1 of the replaced channel credit bandwidth c_cred (S92).

  Further, the channel C1 having the next lowest priority is selected (S96), and the false free bandwidth net_deficit becomes 0 (N in S82), so no further error compensation is necessary, and the channel use bandwidth c_used of the channel C1 Is set to zero (S94).

  In this way, the channel usage bandwidth c_used of a channel having a negative channel credit bandwidth c_credit is set to a negative value equal to the channel credit bandwidth c_credit so that an extra bandwidth can be used. For channels with positive c_credit, it is possible to compensate for the fake free bandwidth net_deficit by using the bandwidth corresponding to the channel credit bandwidth c_credit in order from the channel with the lower priority. .

  Hereinafter, the points not described in the above description will be supplementarily described.

[4] Repeat processing of requests for large size data When accessing large size data such as images and sounds stored in the storage device 150, the size of the requested block is large and does not fit in one time frame. Sometimes. Even if the block size fits in one time frame, if a request with a large block size occupies most of the time frame, other requests cannot be allocated within the time frame, and other requests Response is extremely worse.

  Therefore, the multiplexer 30 processes a request with a large block size by dividing a request with a large block size into requests with a small block size and repeatedly issuing the divided small block size requests. Thereby, it is possible to prevent the response time of other requests from being deteriorated by a request having a large block size.

[5] Disk scheduling The disk scheduling operation by the disk scheduler 20 will be described in detail. The disk scheduler 20 performs disk scheduling in consideration of the access pattern for the disk in order to maximize the throughput or response time performance. The disk scheduler 20 appropriately selects or uses a disk scheduling algorithm. For example, in order to improve the response time, an SSTF algorithm that performs scheduling in ascending order of seek time can be used. In order to improve the service time, an elevator algorithm or a SCAN algorithm can be used.

  When scheduling a request in a frame, the disk scheduler 20 refers to the application type of each request and determines what disk scheduling method should be combined in what ratio. The disk scheduler 20 uses the application type as a disk scheduling hint.

  Application types include multimedia, real-time, interactive, and file transfer. The disk scheduler 20 determines what type of request is multiplexed in the frame and at what rate. Interactive applications are sensitive to response time, and multimedia applications are sensitive to service time. When the ratio of the application sensitive to response time to the application sensitive to service time is α vs. (1-α), the SSTF algorithm suitable for the response time type and the elevator algorithm or SCAN algorithm suitable for the service time Used in combination at a ratio of α to (1-α).

  Further, the disk scheduler 20 may evaluate the request size as one factor, and control so that a request with a smaller request size is serviced earlier. For this purpose, the STEDV (Shortest Seek and Earliest Deadline by Value) algorithm can be improved and used.

  As described above, according to the request scheduler 40 of the present embodiment, the multiplexer 30 is provided in the preceding stage of the disk scheduler 20, and requests from a plurality of connections are multiplexed while guaranteeing the reserved bandwidth of each connection. be able to.

  The multiplexer 30 has two ports, a channel port and an output port 42. As long as the total amount of requests arriving at each channel port does not exceed the reserved bandwidth limit of that channel, the channel port is opened and the channel queue is opened. By bypassing the request directly to the output port 42, the overhead due to queuing to the channel queue can be eliminated, and the channel can be prevented from becoming a bottleneck.

  Also, if the total amount of requests arriving at each channel port exceeds the reserved bandwidth limit for that channel, close the channel port and queue the request in the channel queue until the next time frame begins, The bandwidth used by each channel can be limited.

  Further, when there is no request in the queue of the disk scheduler 20, the free bandwidth port 44 for taking out the request queued in any channel queue and passing it to the disk scheduler 20 is used as a spare port, and the output port 42 Since it is provided separately, free bandwidth can be used effectively. And if the request arrives late and it turns out that it is an error to assume that it is free bandwidth, the bandwidth between channels is compensated by compensating the stripped bandwidth with the next time frame. The fairness of width allocation can be guaranteed.

  Further, according to the present embodiment, the multiplexer 30 multiplexes requests from the connections according to the bandwidth of the disk transfer reserved by each connection, and the disk scheduler 20 sends the disk access request to the disk access pattern. Schedule based on Thereby, when there is a disk access request from an application having various service quality requirements, the service quality requirement of each application can be guaranteed by multiplexing the requests by the multiplexer 30, and at the same time the disk scheduler 20 Disk performance can be maximized by disk scheduling based on access patterns.

  Although the storage device generally operates in a non-deterministic manner (and is not deterministic), it cannot predict when an access request to a disk will be executed. 40, the multiplexer 30 multiplexes the requests of each connection in a time frame processed in a fixed time, and the disk scheduler 20 performs disk scheduling by changing the execution order of requests in the time frame. The time at which the request is executed becomes deterministic, and the quality of service can be guaranteed.

  The present invention has been described based on the embodiments. The embodiments are exemplifications, and it will be understood by those skilled in the art that various modifications can be made to combinations of the respective constituent elements and processing processes, and such modifications are within the scope of the present invention. .

  The channel of the multiplexer 30 can be used for various purposes. A channel may be provided separately for each connection, and one channel may be shared by a plurality of connections. For example, a channel may be prepared for each application type such as real time, multimedia, interactive, etc., and a plurality of connections of the same application type may share a corresponding application type channel. If there are multiple connections for an interactive application, all requests for those connections are queued to the interactive type channel.

  When the request scheduler 40 is used in a multiprocessor system, each processor can establish a connection with the storage device 150. In that case, the channel of the multiplexer 30 may be provided for each processor, or may be provided for each connection established by the processor.

  In the embodiment, the disk scheduling has been described as an example, but the present invention can also be applied to access scheduling for devices other than disks. Further, the multiplexer 30 according to the present embodiment is separated from the disk scheduler 20 and can be applied to fields that require multiplexing processing of requests from a plurality of connections in addition to access scheduling for devices such as disks. For example, the present invention can also be applied to a case where an access request to a server or the like via a network is stored in a FIFO buffer and multiplexed.

It is a figure explaining the structure of the storage subsystem which concerns on embodiment. It is a block diagram of the request scheduler of FIG. FIGS. 3A and 3B are diagrams for explaining in detail the internal configuration of the main channel and the reservation channel and request input / output control of FIG. It is a flowchart which shows the whole flow of the request | requirement multiplexing process by the request scheduler of FIG. FIG. 5 is a flowchart showing processing when event A “request arrival” in FIG. 4 occurs. FIG. FIG. 5B is a flowchart showing in detail a process of updating a waste bandwidth and updating a state of a channel port in FIG. 5A. FIG. 5 is a flowchart showing processing when event B “disk scheduler queue is empty” in FIG. 4 occurs. FIG. FIG. 5 is a flowchart showing processing when an event C “start of a new time frame” in FIG. 4 occurs. FIG. It is a flowchart which shows the detailed procedure of the error compensation process of FIG. It is a figure explaining the procedure of the error compensation process of FIG. 8 using a specific example.

Explanation of symbols

  10 device drivers, 20 disk scheduler, 22 disk scheduler queue, 30 storage multiplexer, 31 main channel, 32 reserved channel, 33, 34 channel port, 35, 36 channel queue, 38 timer, 40 request scheduler, 42 output port, 44 free Bandwidth port, 50 SCSI class driver, 60 storage service provider, 62 connections, 100 storage subsystem, 120 storage interface, 130 file system, 140 operating system, 150 storage device.

Claims (8)

Provided for each service requester, accepting bandwidth reservations for processing requests from each requester, multiple channels through which requests are input from each requester,
An output port for outputting requests passed from a plurality of the channels,
A channel in which a time frame that is a time unit for multiplexing requests from multiple request sources is provided, and at least a part of the reserved bandwidth is intercepted by a request output from another channel in a certain time frame Is subjected to compensation for the pre-empted reserved bandwidth in the next time frame.   When there is no request in the output destination queue of the output port at an arbitrary time point of the time frame, the free bandwidth is allocated to a certain channel on the assumption that the free bandwidth has occurred in the time frame. The multiplexer according to claim 1.   When a priority that is dynamically changed according to the number of requests waiting in the channel is provided for each channel, the channels are selected in descending order of the priority, and the free bandwidth is assigned to the channel. The multiplexer according to claim 2, wherein: is assigned.   Bandwidth that could not be allocated if it was not possible to allocate bandwidth to the request due to a delayed arrival of the request from the channel that reserved the bandwidth in the time frame where it was assumed that free bandwidth occurred. 4. A multiplexer according to claim 2 or 3, wherein the width is counted as a reserved bandwidth intercepted.   For channels that pre-empted the reserved bandwidth of other channels, the bandwidth used in excess of the previous time frame is regarded as already used in the current time frame, and the reserved bandwidth is allocated. A multiplexer according to any one of claims 1 to 4, characterized in that it compensates for a pre-empted reserved bandwidth. A multiplexer that outputs requests input from one or more requesters of the service;
A scheduler for scheduling and executing the request output from the multiplexer,
The multiplexer is
A plurality of channels that are provided for each request source, accept bandwidth reservations for processing requests from each request source, and receive requests from each request source;
An output port for outputting requests passed from a plurality of the channels,
A channel in which a time frame that is a time unit for multiplexing requests from multiple request sources is provided, and at least a part of the reserved bandwidth is intercepted by a request output from another channel in a certain time frame Is an information processing apparatus that receives compensation of a reserved bandwidth obtained in the next time frame. The scheduler includes a scheduler queue for queuing requests output from the output port;
If there is no request in the scheduler queue at any time point in the time frame, the multiplexer assumes that free bandwidth has occurred in the time frame, and if there is a waiting request on a channel, the request The information processing apparatus according to claim 6, wherein the free bandwidth is allocated to the channel by taking out the data and passing it to the scheduler. A channel function that accepts a bandwidth reservation for processing a request for each service request source, and receives a request from each request source through a plurality of channels;
An output port function for providing an output port for outputting requests passed from a plurality of the channels;
A channel in which a time frame that is a time unit for multiplexing requests from multiple request sources is provided, and at least a part of the reserved bandwidth is intercepted by a request output from another channel in a certain time frame On the other hand, a program for causing a computer to realize an error compensation function for compensating for a reserved bandwidth obtained in the next time frame.

Images