TW402700B - Method of allocating the service request - Google Patents

Abstract

A method of allocating the service request with data stored by round robin. Each service request is served by a real server respectively and stores the present data block read now and the the shift data block saved previously respectively by the buffer block series having several buffer blocks. The method includes the following steps. Firstly, decide the server for updating that corresponds to each service request respectively. Secondly, search the service request loop sequence which includes parts of the service requests according to the real server and the server for updating that corresponds to the respective service requests. In the service request loop sequence, one of the real server of the service request is the same as the server for updating of the previous service request, and the server for updating is the same as the real server of the next service request. Finally, change the real server of each service request in the loop sequence to reduce the number of buffer blocks.

Description

V. Description of the invention (1) 4U ^ VU ϋ The present invention relates to service requirements (service ^^ "of a scalable multimedia system (scaUbie system). Configuration is particularly a dynamic service requirement configuration method, which is enough for the system ^ For multimedia system services with low initial delay and low buffer requirement. In multimedia systems, a large bandwidth is generally required to transmit data to a few clients or from hard Read data on discs or other storage media and the system must be expandable (sc a 1 ab i 1 it ty) in order to respond to the increase in the number of service customers. However, when adding hard disks to store richer multimedia data, When the number of customers who need services increases, the initial delay of the client will increase. This is the case for some multimedia systems that require low initial delay characteristics, such as on-demand video systems (V 丨 d⑸

On Demand ’V0D) is a problem that must be solved. θ Real-time characteristics of general systems, the main factors 疋 Startup response time, which is the initial

delay. The shorter the actual delay time, the better the instantaneous characteristics of the system. But as mentioned before, the characteristics of low initial delay will be destroyed with the increase of data volume and customers. At this time, the system must pay extra costs, such as using faster input / output (I / O) memory. The fetching device either uses a faster microprocessor or uses more buffer sizes to speed up the access operation. For example, in 5 Ghandehar i z adeh, S. H. Kim, W. Shi, R

Latency in SPIE Vol.

Zimmermann, " On Minimizing Startup Scalable Continuous Media Servers "

M.27Q0 Five 'Invention Description (2) 3020, 1 997 In this paper, it uses replication data and request migration to achieve load balancing and reduce initial delay. However, the system must use a lot of extra buffers. The following describes the initial delay phenomenon in scalable multimedia systems and the two methods used to solve the initial delay phenomenon in the conventional technology with the diagrams. The first is a data replication method, and the second is Distributed bandwidth (shared bandwidth) method. Figure 1 shows a schematic diagram of a general scalable multimedia system architecture. In FIG. 1, the 'extensible multimedia system includes an allowance / dispatch server (administrate and dispatch server) 30, a plurality of servo devices (21'22), and a plurality of storage devices (i0, 11, 12). The allowance / distribution feeder 30 is responsible for permitting and assigning service requests from clients (not shown), as well as integration and communication coordination among various devices in the system. The feeding device (21, 2 2) reads data from the storage device (丨 〇, 11, 12) in accordance with the customer service requirements assigned by the allowance / distribution server 30. In practical applications In the above, the storage device may be a medium capable of storing data, such as a hard disk, an optical disk. The devices in these systems exchange control signals and data through transmission media 99, such as using

Ethernet > FDDI 'Fast-Ethernet' Gigabit-Ethernet ^ ATM, etc. 'or an input / output bus of SSA, Fiber channel, etc. These servo devices and storage devices in the system can be freely added, reduced or exchanged within the specifications to adjust the overall system performance. In addition, in general scalable multimedia systems,

V. Description of the invention (3) The technology of data stripping, that is, dispersing each multimedia lean material into several data blocks, which are stored on each storage device to achieve load balance (1 oad balance). And each server will take turns reading the data blocks originally stored on each storage device in a circular manner (r0und r0bin). FIG. 2 is a simplified system schematic diagram of an expandable multimedia system, which is used to describe the above-mentioned data decentralized processing and patrol service mode. This simplified system includes four servo devices (labeled s 丨, S2, S3, S4) and four storage devices (labeled HM, HD2, HD3, HD4). In Figure 2, the system uses a hard disk drive as the storage device. Each storage device may have more than one hard disk or a disk array. The small squares in it indicate that it is stored in this storage. Multimedia resources on the device :. The hypothetical situation here is {'provide a movie film as the system's evening media data', and this multimedia data is stored in each storage device in a continuous sequential Round Robln way. In each storage device = the number in the square, it represents the movie data block number (data 2H ^ UenCe nUmber). In detail, the 0th data block is placed on the = storage device and the i-th data area The blocks are placed on the storage device hd2, and so on. In addition, in the second figure, each servo device (S1, s2, δ3 'S4) is eight packets: Four small squares represent service requirements that can accommodate the customer. This ============================================================================================================== =========================================================================================== Four small squares to represent the service requirements of the client to be able to accommodate the second request; and assuming that each serving device is at most at the same time Can serve four requirements. As shown in the figure, the first small square on the feeding device S1 has been filled

V. Description of the invention (4) Above: A, which means that the temple service device is currently assigned to the service request A by transmitting the 0th movie data block. And because the movie film data is placed in a circular manner on the four other devices, so for each servo device, it is stored in the same way as j 'periodically to each storage device ± to read ^ = Rotating material blocks. That is to say, the servo device will have 1 cycle (Avice round, so-called service week, the service device will serve all customers once). At the same time, it will switch to another server 5 to get the required information. The top part in Figure 2 takes the lead. There are two video data blocks, in addition, there are two on the small box; m small arrow, the small arrow pointing to the small box represents feeding clothing ㈣ read the video data block from the stored HD1 'and store it in the buffer; The small arrow pointing to the small box represents the data sent to the client. Therefore, in the next service cycle +, the servo device 51 will switch to the surviving device ship, and continue to read the next (first) data block of the movie, and so on. This assumes that the system uses a single buffer scheme (One Buffer Scheme), and assumes that the amount of data that the server reads the movie will always be sufficient for the client. In the following discussion, a simplified system schematic diagram similar to FIG. 2 is used for explanation. The phenomenon of initial delay will be explained in detail based on Figure 3 and the example shown. Assume that the system is just starting as shown in Figure 2. A new service request A appears in the first service cycle. At this time, the system will

C: \ Program Files \ Patent \ 〇356-3848-E.ptd page 8 40 5. Description of the invention (5) Ask A to configure the temple service device for services to access the storage where the M solid data block is stored. The server S1 is requesting A to read the 0th data area ^. On the storage side, it can help. Then, "Assuming another new service appears in the second service cycle, because the device reads in turn." The storage device is based on the servo device S4, so the system will request service B. Configuration: on servo device S4. At the same time, the verbal device. Then change to the storage device HD2, and request A to read the i-th data block for the service. In the next few service cycles, service requests C and D were received in the 5th service week, service request E was received in the 7th service cycle, and service request F was received in the 9th service cycle. . Among them, each service request is a servo device that is configured to read the storage device 0 at that time. 2 Γ Figure 1, circle 13: Figure 3 shows that the system is not intended to be simplified during the 9th service cycle. As shown in the figure, there is a service request A, c, D, and F in the servo device S1, a service request E in the servo device S3, and a service request B in the service request S4, and there is no service request in the servo device S2. . Under the above-mentioned permission / distribution control mechanism, when a service request occurs, the system will give priority to the servo device that is currently accessing the storage charge stored in the 0th block. However, this permission / distribution control-control mechanism is still acceptable at the beginning of operation, but after a period of operation, the report easily causes the service request to be concentrated on a specific servo device, such as the server S1; or A servo device has a low utilization rate, such as the servo device S2, which is a case of unbalanced load. The phenomenon of initial delay can also be explained according to Figure 3. If there is any service requirement

Five 'invention descriptions (6) ^ ^ / 0 y a' is added at the 9th service cycle 'or is added during the service cycle when the servo device s 1 is rotated to the health storage device HD 1, because the servo device s丨 The service request that can be received is full. The system cannot be directly configured on the servo device S1. At this time, the system will delay the time to start the service request. Wait until the server device with sufficient service capacity can access the storage. This new service request is only started when the device HD1 is on, so it also causes the initial delay of the new service request, and this situation is more likely to occur especially when the system is approaching full load. Therefore, there are two problems with traditional multimedia systems. The first is load imbalance and the second is initial delay. At present, there are two methods to solve the load balancing and initial delay. One is to use data replication. The first few data blocks of the video are copied to each storage device in advance. The other is to use the additional frequency of some servo devices. Wide to assist in processing the movie to start reading several data blocks. The contents of these two conventional techniques will be described below with reference to the drawings. Fig. 4 shows a simplified diagram of the system in the case where the system uses a data copying method. As shown in the figure ', in the storage devices HD2, HD3, and HD4, there are stored some additional data blocks, respectively. The 0th data block is stored in the storage device HD2 'The 0th and 1st data blocks are stored in the storage device HD3' The 0th, 1st, and 2nd data blocks are stored in the storage device HD4 Data block. That is, each storage device (except HD1) separately stores the first few data blocks in the same movie data.

When the system receives the service request A at the beginning, it will be configured in the server S1 for service. If there is another service request B at this time

C: \ ProgramFiles \ Patent \ 0356-3848-E.ptd page 1〇

In the example, the server device S2 or the servo device S3 is used, and the spleen server I4 is used on the device S4. In the system situation shown in Figure 4 and Figure 2, set ^ = on .: If the customer is identified according to the 0th data block of the data, the movie S2 cannot be read immediately. At this time, the 0th data is retained. Block, so the servo device has one data block. The other S it simultaneously obtains the first and first service Rongbing R planes. Under the extended solution, the system must be a service that can accommodate two data blocks. Enter the next buffer # il 1: Servo device 52 will rotate to the storage device HD3 when the month is full. At this time, the feeding device s2 only needs to serve the second data of Hongling 嫂 i. Block, and at the same time a service request B is currently consuming the 0th data area 'ghost, so the = and 2 data blocks stored in the buffer, and the rest of the service cycle can also be deduced by analogy. In the first method, the first few data items of the movie are copied in advance to each storage device. Therefore, the system can not only allow the addition of service fees immediately, but also achieve load balancing. However, this method has three disadvantages. First, every storage device must You need to spend extra-some storage space to store = some data blocks at the beginning of the data. Second, the system must provide an extra $ buffer space to place shell material blocks that may need to be read out in advance for service requirements. The third 'servo The device needs to read more than one block of data for a single service when it starts to service a new service, which means that the server needs to perform a burst of data retrieval (burst data retrieve), which is naturally bad for the system. influences.

C: \ ProgramFUes \ Patent \ 0356-3848-E.ptd page 11

V. Description of the invention (8) 4ΤΓ27ΙΓ0 '一' ~ --- Then, according to Fig. 5, the second distributed bandwidth method to achieve system load balance and initial delay will be described. The decentralized bandwidth method is mainly used at the beginning of the service request, using several servo devices to perform the reading operation at the same time 'to achieve the purpose of low initial delay. Referring to FIG. 5, when the system receives the service request A at the beginning, it will be configured in the servo device S1 for service. If another service request B joins at this time, according to the principle of load balancing, the system will configure service request B in feed clothing S2. However, since the servo device S2 currently serves the storage device HD2, and 1 does not include the 0th data block, the system initially places a temporary service request b in the servo device S1. As a result, the servo device si and the servo device S2 will complete reading of the 0th and 1st data blocks in the same service cycle. In the next service cycle, the temporary service request b in the servo device S1 will be canceled, and the servo device S2 will tour the service to the storage device HD3 and continue the service. Same as the previous method $, the system also needs to use a certain number of buffers to temporarily store these pre-exported data blocks. ° Using distributed bandwidth mode ’also allows the system to achieve load balancing and low initial latency. Compared with the previous data copying method, not only does it not require additional storage space to store the beginning of the video data, it also does not cause the instantaneous reading demand of a particular server. 'It may cause unstable situations when the system load is fragmented. The problem. However, like the data copy method, the system must pay a large number of buffers for some service requests to store those data blocks that are read in advance. In summary, although data replication or distributed bandwidth can solve

V. Description of the invention (9) Load balancing and the problem, that is, block, provide additional data blocks with the premise of considering the requirements and delay of deployment according to the above method. This method can be stored in a cyclic manner 'for rotating blocks. Each service set performs a service, and is arranged in order to divide the current data block. The condition of the above service request device is the required offset when the required offset service corresponds to the actual sub-service request. Update the requirements of the server installation service

Low initial delay is a necessary buffer area of the system. The method of the present invention can effectively and effectively flush the area. For the purpose of this, this applies to a series of data recovery services. The above storage requirements are respectively and correspond to other storage through and Correspondence of pre-storage when the number of service required blocks is shifted. The data block server device and service request are the same as the service request. The problem with the servo device is introduced, but another new request for pre-storage for some service will be introduced. The main purpose of the data area is to reduce the number of inventory storage devices to the above-mentioned actual offset asset update requests in the maintenance and reduction of the invention. The to-be-updated circle sequence is internationally servoed and it remains the same. The purpose is to provide a system that supports the overall performance of the system load balancing technology. A service requires that it includes a plurality of storage material blocks, and that one of the above-mentioned service devices is sequentially read and one buffer server is installed. Material block. Server, update the servo from the actual next, the root servo. In terms of the service device and the previous block, first, the above-mentioned devices to be included in the servo data loading service. The search request will be returned to the service as well as an additional provision at the beginning. The device on the deployment side will be equipped with the servo data buffer area. The read before the block decides to update the service line when the service request is separated. The update server that contains the service sequence and the service request is separated from the last one. For each search,

C: \ ProgramFiles \ PatentA0356-3848-E.ptd page 13 V. Service requirements of invention description (10) Each service setting in the loop sequence is set to its corresponding feed clothing to be updated. Will, the actual servo installed buffer The number of zones becomes smaller, so it can be reduced to 5. Since the overall needs are changed due to the change of the servo device according to the requirements of maintaining load balance and low initial delay, it can also be added. A service request is when the service request is entered into the device by the temple service device to be updated. If the number of blocks is zero, the best system can be achieved. This best effort is in the buffer “string” corresponding to the service requirements; two sets of “storage blocks” are used. In addition, the information about the current use of the magic armor order can also include a setting step to find the actual temple clothing; 5 ϋΦ cylinder According to each service or not: the "international" server device and the servo device to be updated, set the pair: find and point the actual servo device to the directed edge of the server device to be updated. = In the search step, the required service request loop sequence can be searched according to whether these directed edges constitute a directed cycle. If there are several service requirements, all of which correspond to the same actual feeding device and the servo device to be updated f ', then a single directed edge can be used to represent these service requirements to achieve the purpose of simplifying the directed graph. When a new service is requested to join, the corresponding actual servo device is determined according to the load balancing law. After determining the server device to be updated corresponding to the new service request, a service request loop sequence including the new service request may be searched out. When new service requirements are included in different service request loop sequences, the maximum offset data can be reduced according to the adjustment steps C: \ ProgramFiles \ Patent \ 0356-3848-E.ptd page 14

Law of the number of data blocks That is, if a sequence containing more than one of the new ones is found, the rule sequence of the number of blocks is found. Determine the service requirements that need to be adjusted. If a new service request is added, the number of service request loop sequences in the service step in the search step exceeds the maximum offset amount that can be reduced after the adjustment step. A simple description of the loop diagram of the service requirements: ", features and advantages can be more obvious and easy to understand 'and in conjunction with the accompanying drawings, a detailed description. To make the above-mentioned objectives of the present invention a preferred embodiment is given below, : The display is a schematic diagram of a general scalable multimedia system architecture. In the extensible multimedia system, Huanfang is used to explain the simplified schematic diagram of the system that stores data in a cyclical manner. Figure 3 shows a simplified diagram of the system used to explain the initial delay phenomenon. Data replication method to resolve load contention Figure 4 shows a simplified schematic diagram of the system using the balance problem in the conventional technology. Schematic diagram of bandwidth structure to solve load structure. System of Directed Graph Structure Figure 5 shows a simplified schematic diagram of a system using a distributed balance problem in the conventional technique. Fig. 6 shows a directed edge in the embodiment of the present invention. Fig. 7 shows a simplified schematic diagram in the embodiment of the present invention. FIG. 8 shows a directed graph corresponding to FIG. 7. Figure 9 shows the first example of the embodiment of the present invention

V. Simplified schematic diagram of the system of the invention description (12). FIG. 10 shows a simplified schematic diagram of a system according to the first embodiment of the present invention.丨 The 11th diagram of the consumption example after the transfer process is required is not as simple as the present invention. A simplified system schematic diagram of the first example is shown before the transfer process is required. The second example does not intend to simplify the system after the transfer process. The 13th diagram is not a simplified diagram of the present invention. The system is simplified. Example: The 14th diagram showing the search loop procedure is shown in the 13th diagram. The simplified diagram of the system after the simplified processing. Figure 15 shows the search tree diagram for deep optimal pure search without using Figure 13. Figure 16 shows the relationship between the total service requirements and the number of buffers in the simulation system. Description: 10, 11, 12 ~ storage device; 21, 22 ~ servo device; 30 ~ allow / distribute feeders; 99 ~ transmission medium; S1-S4-servo device; HD and HD4 ~ storage device; 40 ~ means actual The nodes of the servo device; 4 indicates the directed edge of the service request; 42 ~ indicates the nodes of the servo device to be updated. Embodiment: The service request configuration processing method disclosed in the present invention is mainly used for -------- ----• One —-— Less Qiao number of service requirements needed him buffers, and the means is. Ring to \ requirements transitive (cyclic request migration) way to reach ... In detail

\ Patent \ 0356-3848-E.ptd page 16 5. Description of the invention (13) 'If the requirements of 4 other feeding devices are changed and arranged in each servo device, you can use the method of recursive request to Reach the present invention. In other words, °, Hei Yi _ ;, the original service requirements of the feeding device cannot be changed during the load balancing of the system, and must be moved into a service request at the same time. , Move into the service requirements during the process. Service requirements' can not be transferred in directed =. In the embodiment, the concepts of directed edges (directed coffee) and: 1 recte (1 graph) are used to determine a set of services that can perform requirements = Claim. First explain the meaning of the directed g element in this embodiment. (1) Node: indicates a point in a directed graph, which corresponds to a server device in the system; (2) Directed edge: indicates that it exists in a directed graph. The directed edge between two nodes 'corresponds to the service request that the system has received' whose direction is the node corresponding to the actual servo device (described later) of this service request, pointing to the server device to be updated for this service request ( Nodes described later). Figure 6 shows a schematic of a directed edge structure. In Fig. 6, the nodes (40, 42) at both ends represent the actual servo device of the service request A (41) and the servo device to be updated. The situations shown in the figure are the servo devices S1 and S3, respectively. The actual servo device refers to the servo device that was originally allocated to the service request a by the system. The server device to be updated is another server device in the system, which means that if the service request A is changed to be served by the server device to be updated

C: \ ProgramFiles \ Patent \ 0356-3848-E.ptd page Π 5. Description of the invention Q4) = Yes, the number of punching areas will be lower than the number of buffers required by the actual servo device for service. When all service requirements in the system are represented by individual directed edges, a directed graph corresponding to the system can be constructed. The following describes the concept of directed edges and directed graphs based on the system service request status shown in Figure 7. The symbols used in Figure 7 are the same as those in Figures 2 to 5 above, and are not repeated here. In FIG. 7, the servo device S1 includes a service request A and a service request E, the servo device S2 includes a service request j, the service device S3 includes a service request c, and the feeding device S4 includes a service request \. Each service request is configured with 2 to 3 buffers. In each buffer and series, the leftmost represents the data area consumed (or used) by the current client, such as the second data in service request A. Block; the rightmost represents the data area & read by the actual word server, such as the fourth data block of service request A. Take service request A as an example. Service request A is currently served by a servo device. In the current service cycle, the service request A 疋 reads the * th data stored in the storage device Hm through the feeding device 81. At the same time, the client is using the second data read into the buffer in advance. Shell material block. In other words, the buffer queue corresponding to the service requester contains two buffers, one of which is the buffer being accessed by the client, namely the Λ ', and the other two are the buffers stored in advance (including the first 3 and 4 data blocks). The following are the possible pending servo devices corresponding to the service request A, including ί Π = 2 is the servo device to be updated, which is set in the servo device ’The number of punching areas changed from 3 to 2’ This is because no storage is required

Five 'invention description (15) buffer area for storing the fourth data block. The second type is to be changed in the servo device S3, which is an optimized situation at this time. It can be changed from 3 to ^ ', that is, it does not need to store any pre-quantity data block. In this optimization case, the servo server to be updated is: according to the service request, the data block currently being consumed, β is the storage device currently accessed by that servo device, and this It is the required servo device to be updated. In this embodiment, the main situation a and optimization are described, but for those who are familiar with this skill, + if the servo device to be updated is selected according to actual needs. Therefore, the service = I root will correspond to the directed edge pointed by the servo device S1 to the servo device S3. Request A According to the same processing mode as the above service request A, 4M can decide that it should be on the directed edge of other service requests. As shown in Figure 8, the service request B corresponds to the servo device S2 pointing to the servo. For device Si = directed edge, service request C corresponds to directed edge from servo device S3 to servo device Sl, and service request D corresponds to directed edge directed from servo device S4 to servo device $. Service request E corresponds to Servo device S1 points to the directed edge of S4. ° Clothing According to a directed graph similar to that in Figure 8, you can directly look at the house. π m mother and child service requirements can be performed under conditions that require recursive processing. Basically, Cui_has. Edges do not constitute a requirement for transfer. As mentioned earlier, when you move forward and move forward, in addition to the effect of reducing the number of buffers, you also need to maintain the original load balance of the system. Therefore, in this embodiment, σ ^ w / is used to calculate the service requirements that constitute the service request loop sequence, and the requirements are shifted in order to maintain load balance. A service requires the loop sequence to be a _loop on a directed graph

c: \ Program Files \ Patent \ 0356-3848-E.ptd page 19 5. Description of the invention (16) 40 ^ 700 'form, for example, Figure 8 contains the service requirements a, c loop and package service requirements B, E, D loops. In the compliant service request loop sequence, the actual servo device for each service request will be the same as the server to be updated for the previous service; on the other hand, each new service device for each service request is also the same as the next service request. The actual servos are the same. When we find such a service request loop sequence, we can change the server device to be updated for each service request to the actual server device, that is, transfer the server device where each service request is located. Thereby, the total number of buffers required can be reduced, and the object of the present invention is achieved. Following the two loops shown in Figure 8, the following two examples are used to illustrate the process of requesting recursion. First example: Figure 9 shows the simplified system schematic diagram of the service request loop sequence (including service requirements A and C) before the request relocation process in the first example, and Figure 丨 the system after the request relocation process Simplified diagram. As shown in FIG. 9 ', the current service request A is served by the server S1, and the service request C is served by the server S3. The two service requests currently use three buffers each. In addition, it can be found from FIG. 9 that the service request a is currently consuming the second data block stored on the storage device HD3, and the current storage device HD3 is being accessed by the servo device S3. According to the above definition, the temple can be determined The serving device S3 is a feeding device to be updated with service request A. On the other hand, the service request C is currently consuming the 0th data block stored on the storage device hdi, and the current storage device HD1 is being accessed by the servo device S1. According to the above definition, the servo device S1 can be determined as a service Requirement c

C: \ ProgramFiles \ Patent \ 0356-3848-E. Ptd page 20

V. Description of the invention (17) The server to be updated from 40 to 200 has a direction. Therefore, if the actual servo cycle of the international servo device is rotated to the storage client, the current buffer can be changed to the storage client. Currently six buffered condition service devices are available in the buffer zone. At this time, it corresponds to the loop on the graph. In the current service cycle, the ‘change setting is set to servo device’ is changed to set the server as shown in FIG. 10, the device is stored in HD4, and the third data block being consumed is read. On the other hand, the service requires HD2 to read the first data block consumed by the data. Through the above-mentioned data delivery block, the reduction was reduced to two unchanged. For the directed edges of service request A and C, the service request A can be implemented with S3 and the service request c can be served with device S1. In the next service, the servo device S3 data block of A is required, and this is also the same, so only a single servo device S1 that needs to find c will rotate the block, and because this is exactly the customer, the whole need only be a single A mobile operation can load the originally required data blocks while the system loads the second example: The second example differs from the first example in the number of service requests in the loop. Figure 11 shows the simplified system schematic diagram of the service request loop sequence (including service requirements B, D, and E) before the request relocation process in the second example, and Figure 12 is the simplified system schematic diagram after the request relocation process. . As shown in FIG. 11, the current service requirements B, d, and E are serviced by the servo devices S2, S4, and S1, respectively. At the same time, 2, 3, and 2 buffers are used. Based on the data blocks being consumed by the clients required by each service, it can be determined that the corresponding servo devices to be updated are S1, S2, and S4, respectively. Similarly, the directed edges corresponding to the service requirements B, D, E will constitute a return on the directed graph.

C: \ ProgramFiles \ Patent \ 0356-3848-E. Ptd page 21 402700

V. Description of the invention (18) circle. Therefore, if the service requirements B, D, and E are recursively requested during the current service cycle, and set to the corresponding server to be updated, then in the next: service cycle, as shown in Figure 12 Happening. The new servo device si of service request b will be rotated to the storage device HD2 and read the fifth data block; the new servo device S2 of service ^ for D will be rotated to the storage device hd3 and read 3 VI. The new servo device S4 of block two service request E will rotate to the storage device Γ to take the fourth data block. These are the data blocks that the client is consuming, so only a single buffer is needed. Through the above-mentioned data transfer operation, “the seven data blocks originally required can be reduced to two data blocks” and the conditions for system load balancing have not changed. According to the above, it can be known that the number of buffers can be reduced through the method of requesting shifts, while taking into account load balancing. The only remaining question is how to find the loops on the directed graph, because if the system has a lot of service requirements, it is conceivable that the system must pay extra computing time to find the loops in the directed graph. This will have a considerable impact on the initial stage, so it is necessary to reduce its complexity in order to ensure that the system service has the characteristics of low f latency. ° When a new service is requested to join, for load balancing, there are several possible servo devices that meet the requirements for serving this service. The system will configure an actual servo device to the newly added service request. On the other hand, ' Because the storage device for the 0th data block of the movie is indeed — and the only one (without the use of data replication), and in the service cycle that the service wants to focus on, the 0th data area is read Block storage ~

4Q27Q0 V. Description of the invention (19) The server to be installed is also determined, and the device can be determined upon joining. Use a depth-first principle. As mentioned earlier, when the system allows / administrates / dispatch s, it immediately configures its actual servo device as known. Then it can pass through the circle, that is, by adding a new service request to The actual servo device designated by the system is the starting point of the servo device to be updated, and the direction is composed of directed road control (the addition of the direc service request will form a rational 'so that all the services in the loop will be 0. In addition, this embodiment also Simultaneous graphs. First, the directed graph only records service requirements', that is, there is a service that uses only one buffer to pre-store and can contain thousands or tens of thousands of the same service requirements but present the same. It indicates to the edge that the complex service request can be deleted on the edge when a certain loop is eliminated, and if the directed edge formed only has this, the actual servo device and the servo to be updated are based on this relationship. In this embodiment, st) The search method to search for the allocation / server erver) will receive the new service request device, in other words' At this time the actual servo is too deep-first search party To find the back-to-be-updated servo device as the root and set it as the end point, if you can find a path from the actual servo device as the end point (there is a ted path), then you can know that this new service moves to the loop 'reuse request. All the requirements can be achieved by reducing the buffer. Other methods are used to simplify the offset buffer of the server that currently uses more than two buffers, and there is no need to record the request. Second, in complex systems, serviceability is required, so it is very likely that the situation will not go to the side. At this time, this actual example is a service request, and when a new service is added, one of the service requirements is in the original directed graph. When the service is a service request, the directed service is cancelled.

Description of the invention (20) Fig. 13 shows that there are service requirements A ~ p in the existing feeding and feeding device SH4 in a certain system, and some directed edges, for example, the servo device S4 refers to the directed graph in Fig. 4 Finding C, D, and N can be simplified into a service of the service device S1, which means a simplified directed graph. ",", Representative, Figure 14 at this time. "Suppose a service requires Q to prepare a load balance history, as shown in fW + tube to join the service. If the actual servo is configured according to the ten-balance principle, the service cycle β, Servo device S4 is serving Is2, and it is assumed here that the storage of the first data block in the deployment n service is root. After two toast service devices are S4), then the word service device == priority search until it is found. The actual word clothing court is configured as a taxi. Figure 15 shows the state of the search tree for this depth-first search. First search from @ 服装 置 S4 to the feeding device Slit (directed edge is required for services C, D, N) 'but from the servo device there is no other directed Edge finger, so the search for this part was stopped. Then search from servo device S4 to servo device S2 (directed edges are service requests E, j, p). At this point, the actual feeding device S2 has been reached, so this search is completed, as shown in Figures 14 and 15. Since the newly added service request q and the original service request E,, P will form a circle ', the number of buffers required can be reduced through the processing of the above-mentioned request transfer.

In addition, although the above description is based on the principle of depth first to find the loop formed by the newly added service request in the directed graph, it is not intended to limit its scope. In the application, other search rules can also be used to determine the actual servo device that negatively joins the service request. It can even find several I

C: \ ProgramFiles \ Patent \ 0356-3848-E.ptd page 24 -402700 ______ Five 'invention description (21)-, the case of forming a loop ’Then select the loop in which the maximum number of buffers is reduced. However, too many calculations may lead to a longer delay time. Therefore, in actual and practical applications, the party to find the loop can be determined according to the operating environment and needs. According to the above, the advantage of the present invention is that the load balance is maintained. Can greatly reduce the number of system buffers. The following uses an example to illustrate the effect that can be achieved by requesting recursion. In this example, the total length of the movie is 36,000 service cycles, that is, it is divided into 3600 pieces of data ^ and stored in the storage device. Assume that the number of servo devices is six, and the service requirements from the client have a poisson distribution, and the relationship between the total number of service requirements and the number of buffers is shown in Figure 6-6. In Fig.6, a indicates the situation when distributed bandwidth processing is used but the required retransmission has not yet been performed, and b indicates the case after the required retransmission has been performed. As can be seen from the figure, when the number of service requirements is not high, the difference between the two is not large, but when the number of service requirements increases, the number of buffers required in the two cases will obviously differ. Therefore, it can be said that the system corresponding to the high load of the present invention is quite excellent. Although the present invention is based on a preferred embodiment, the present invention can be used within any skill and scope of the present invention. The patent scope is disclosed as above, but it is not intended to limit the scope of the invention without departing from the spirit and decoration of the present invention. Therefore, the protection defined by the present invention shall prevail.

Claims (1)

6. Scope of Patent Application 1. A service request configuration method can be applied to a system that is stored on a circular basis. The above system includes a plurality of storage devices for storing a plurality of data blocks of a data in a cyclic manner, and at least A servo device is used to access the storage device in turn, and sequentially read the data ^ blocks; each data access service request is served by a second servo device among the above-mentioned servo devices, and corresponds to including at least one buffer. The buffer block sequence of the block is used to store the current data block that is being consumed and the offset data block that is read in advance, which includes the following steps: Determine the server device to be updated corresponding to the above service request, where The above service requires that the number of offset data blocks required by the above-mentioned servo device to perform the service is lower than the number of offset asset blocks required by the above-mentioned actual servo device; according to the actual service device corresponding to the above service request, respectively And the server device to be updated, and search for a service request loop sequence including part of the above service requests. The actual servo device of a service request in the service request loop sequence is the same as the server device to be updated of the previous service request, and the service server to be updated is the same as the actual server device of the latter service request; and the above service request loop is adjusted Each actual service device required by each service in the sequence becomes its corresponding servo device to be updated. 2. The service request configuration method described in item 1 of the scope of patent application, wherein each of the service devices to be updated corresponding to the service request is an offset required when the above service request is served by the server to be updated. The number of data blocks is zero. 3. The service request configuration method described in item 2 of the scope of patent application, C: \ Program Files \ Patent \ 〇356-3848-E.ptd page 26 6. The storage device that is being accessed by the patent application park, and the service request configuration method currently listed in the list is stored. The above servo devices are each based on the actual server device of the node table and the service request to be updated. The corresponding to the servo device to be updated is a loop sequence according to the above-mentioned directed edge service. The service request configuration method described above and the service request configuration method of the server device to be updated. The new server device is the required offset data for the above-mentioned service. The above service requires material blocks. Ge Chong & block of which V: including the first scope of the patent application scope which also includes-setting steps, each new 'and each new service device according to each service request, setting corresponding to the above service server Node refers to the directed edge of the node. In the above search, whether the search constitutes a directed loop, and the above service is found. 5. As described in item 4 of the scope of patent application, which corresponds to the same actual servo device requirement, it consists of a single directed edge. Means. As mentioned in item 4 of the scope of patent application, the waiting service corresponding to the parent service request requires the number of blocks to be updated by the servo device to be updated to be zero. 7. The service request configuration method as described in item 1 of the scope of patent application, which further includes an adding step for adding a new service request, and determining the above-mentioned actual servo device according to a load balancing rule; in the above-mentioned determining step To determine the servo device to be updated corresponding to the new service request; in the above search step, according to a search rule, the search is performed with the to-be-updated servo device as the root, and the actual servo device as the end point to search to determine that the new service is included The requested service requires a loop sequence. C: \ ProgramFiles \ Patent \ 0356-3848-E.ptd page 27 6. Application for special secrets — 8 · The service request configuration method described in item 7 of the scope of patent application, where in the above search step, search out The number of service request loop queues containing the above-mentioned new service requirements is based on a rule that determines the service request loop sequence for the above adjustment steps according to the rule that the maximum number of offset data blocks can be reduced after the adjustment steps. 9. According to the service request configuration method corresponding to each of the service requirements in item 8 of the scope of the patent application, the service request is configured by the server to be updated, which means that the number of service blocks is zero. By.仃 The required offset data when serving C: \ ProgramFiles \ Patent \ 0356-3848-E. Ptd page 28