JPH0970018A - File server - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow a terminal equipment to reproduce a video image with excellent quality by allowing the file server to read data from a disk in an amount larger than the requested amount from the terminal equipment and to store the data into a large capacity buffer and extracting the requested data from the buffer requested for a succeeding disk read, thereby suppressing access frequency to the magnetic disk and improving the read efficiency. SOLUTION: When a read request comes first to a buffer processing section 111 of a file server 101 from a concerned user terminal, a delivery processing section 201 provides an output of a disk read request with an amount more than that request from the terminal equipment to a disk scheduler section 141. The scheduler 141 reads the requested amount of data from a magnetic disk device 151 and stores the data to buffer sections 221, 222. The delivery processing section 201 reads the requested amount of data from the terminal equipment from the buffer and sends the data to a network 401. After that, when the read request is made from the user, the data are extracted from the buffer sections 221, 222 and sends the data to the network 401. Thus, number of times of disk read is reduced and data are sent in a fast response.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention stores digital data,
Regarding the file server to deliver.

[0002]

2. Description of the Related Art Conventionally, a system for allowing a plurality of terminals to share data on a disk of one computer has been known as SUN.
There is a network file system (NFS) proposed by MicroSystem. This consists of a file server that provides data in the storage device, a plurality of user terminals that use the data, and a communication procedure that links them. At present, a magnetic disk or a magneto-optical disk is suitable in terms of cost as a data storage means in the file server. Using this system, multiple computers can use the text data and execution binary on the magnetic disk of the file server.

The server requests from each user terminal by designating the file name, offset, and size of the data on the disk. When the file server receives the read request, it reads the corresponding data from the disk and sends it to the user terminal via the network. In the network file system, in order to optimize the transmission efficiency, even if the application on the user terminal tries to read a large amount of data from the file server, the read request from the terminal to the file server is finely divided and repeated. Will be issued. When an application on the terminal requests a large amount of data from the server, the terminal repeatedly requests the file server multiple times. On the file server,
With respect to the read requests that are finely divided, the disk is read each time unless the cache on the disk is hit.

On the terminal side, the data of the file server is provided to the application on the terminal through the same interface as the data of its own disk, so that the user terminal can access the disk of the file server as if it were accessing its own disk. It is possible to access the data.

The number of terminals that can be connected to the file server is determined by setting in advance. Although there is no structural limitation, the response to each terminal becomes worse as the number of connected terminals increases.

[0006]

There are many applications that operate by reading digital video on a CD-ROM. By storing and sharing the digital video data on this CD-ROM in one file server, it is not necessary to have a large amount of video data in each terminal,
This will lead to lower storage costs. However, since network file systems have traditionally assumed that text data and binary data are shared, the size of video data is generally large, and video data has a time axis limitation when data is transferred. Sharing causes some problems.

When a terminal requests a large amount of data such as video data, disk read is repeatedly issued in the file server. Therefore, when the file server delivers the video data to a plurality of user terminals, a lot of time is spent on the disk read, and the disk read becomes a bottleneck. Therefore, the number of terminals that can be simultaneously connected to the file server as the video server is reduced.

[0008] Generally, a terminal that reproduces video is
If the requested data cannot be obtained within the limited time, the correct video cannot be reproduced. Conventional file servers do not assume delivery of video data, so when the number of terminals connected to the file server increases and disk access increases, the response to requests from each terminal deteriorates. As a result, the video data displayed on the terminal is disturbed.

In the conventional file server, the number of terminals that can be connected must be decided in advance. But,
In the case of video delivery, the number of terminals that can be delivered depends on whether there is enough disk read. If you decide in advance, you can refuse the connection of the terminal even if there is enough disk space, or conversely connect the terminal even if there is no disk space, which disturbs delivery to other terminals. There is a possibility that it will end up.

[0010]

[Means for Solving the Problems] When a terminal requests data from a server for the purpose of displaying an image, after a first read request from the terminal to the file server, a repetitive and continuous request comes periodically. Can be predicted. Therefore, the file server reads the disk more than the amount requested by the terminal and stores it in a large capacity buffer. For the subsequent disk read, the requested data is fetched from the buffer and sent to the user terminal. As a result, the access frequency of the magnetic disk can be suppressed and the reading efficiency can be improved.

Further, when the data in the buffer is sequentially taken out and the data is sent out, the disk read is performed again, but since the data must be continuously sent during the disk read, the buffer is divided into two. Use and use alternately. When the data on one side is used up, the data is sent from the buffer on the other side to the network while the disk is read on that side. As a result, the delivery to the terminal can always be performed from the buffer in the high-speed storage area, and the quick response can be guaranteed.

When video data read requests from user terminals are continuously received, the file server must periodically replenish the large capacity buffer with video data.
A disk read for replenishment must be done before the other side is exhausted while it is in use.
In other words, the disk read deadline is set for each different delivery process. Since the data is different for each terminal, the bit rate is different, and as a result, the time until the deadline is different for each delivery process. When a plurality of delivery units are operating, the scheduler device manages the order of disk reads so that the disk reads are performed in such an order that the deadline of delivery to all terminals is observed. As a result, the disk read can be completed before the delivery of the data to the buffer unit starts.

However, buffering is performed on the terminal side,
There is a possibility that you will lead to some extent collectively. At that time, the buffer in the file server will be exhausted quickly, so we will deal with it by shortening the time until the deadline of disk read by some percentage.

Further, the terminal side may request video data for purposes other than displaying. In that case, the data bit rate is ignored, and the data is requested at a high bit rate. On the other hand, the delivery processing unit of the file server multiplies the ratio of the disk read interval determined by the bit rate of the data from the time when the buffer unit is used up, even if the data in the buffer unit is depleted quickly. Only after a while, read the disc. As a result, it is possible to prevent the disk read from being intensively performed for one terminal and to prevent the delivery to another terminal.

In the file server, when the disk read reaches the limit, the delivery capacity reaches the limit. The amount of disk read per unit time is determined by the total delivery bit rate of the file server. Therefore, the maximum delivery bit rate is obtained in advance in the file server, and when a new request comes from the terminal, whether the delivery is possible by adding the bit rate of the newly requested data to the current total bit rate. You can look up. If it is determined that it is impossible, the read request from the user terminal is rejected. If it is determined that it is possible, the delivery device actually reads the disk and delivers it to the user terminal. By this processing, it is possible to avoid the fact that the delivery to a new terminal is exceeded and the capacity of the disk read is exceeded and the delivery to a terminal already connected is disturbed.

[0016]

In the file server, when the video data is delivered, the access frequency of the magnetic disk can be suppressed and the reading efficiency can be improved. Further, the delivery to the terminal can always be performed from the high-speed buffer section on the main memory. as a result,
It is possible to guarantee a quick response to a request from the terminal.

[0017]

FIG. 4 shows an example of a network environment to which the present invention is applied. Network 401 is often a local area network (LAN) and user terminal devices 421, 422, 423. . . Shall support network file systems. It is possible to apply the file server 101 of the present invention to this environment.

FIG. 3 is a block diagram of the file server of the present invention. It is assumed that the video data is stored in the magnetic disk device 151. With this configuration, even while the data is read from the disk and stored in the main storage device 321,
It is possible to read from another part of the main memory and send it to the network device via the bus.

FIG. 1 is a functional block diagram of the file server of the present invention. The file server 101 includes a magnetic disk device 151, a disk scheduler unit 141, a terminal command distribution unit 131, a network interface 121,
And terminal-compatible buffer processing units 111, 112, 113,
114. . . Consists of The file server has terminal-corresponding buffer processing units corresponding to the number of connected user terminals.

FIG. 2 is a block diagram of the buffer processing unit corresponding to the terminal. It comprises a first buffer unit 221, a second buffer unit 222, a delivery processing unit 201, and a disk read request issuing unit 211.

When a read request comes from the user terminal for the first time, the delivery processing unit performs a-byte disk read from the storage device, and temporarily stores the read data in the buffer unit in the main memory. If the data requested by the terminal is video data, there is a high possibility that the read request will be continuously repeated thereafter. Therefore, a is set to a value larger than the number of bytes requested by the user for read. After that, when a read request is issued from the user terminal, the number of disk reads can be greatly reduced by taking out from the buffer and sending it, and the overhead due to the disk read can be reduced.

Even if the delivery processing unit attempts to deliver the data designated by the read request from the user terminal from the buffer, the required data may not be present in the buffer.
There are two possibilities other than the first read request. One is when the buffer is continuously consumed and used up. At this time, you have to replenish, so prepare another side of the buffer so that you can continue sending during replenishment. At the time of the first read, the read of two sheets is performed, and thereafter, the disk is read each time one sheet is used, and the delivery from the other side to the terminal is performed. The other case where the specified data does not exist in the buffer for the read request is when the read request from the user does not exist in the two buffers because it is not continuous with the previous request. .
In this case, read again from the first two buffers.

FIG. 5 shows a procedure in which the delivery processing unit alternately uses the two buffers and sends the data to the user after the data read request is received from the user terminal. This is triggered by the first read request from the terminal. It is assumed that a read command with offset x and size s is received from a terminal to a file. On the other hand, the delivery processing unit issues a disk read request of size a to the disk scheduler twice. a is set to a value larger than s.
By prefetching in this way, the overhead of disk read can be reduced.

There can be four cases of processing. The first case is when all the data to be read is in the buffer X. Reference numeral 509 in the flow chart shows the processing at this time. In the second case, the first half of the data to be read is in the buffer X and the second half is in the buffer Y. At that time, it becomes like 511 to 515 in the flow chart. In the third case, the data to be read does not exist in the buffer X at all, but exists in the buffer Y at all. At that time, it becomes like 516 to 519 in the flow chart.

FIGS. 6 to 8 show buffers in each of the above three cases. In the fourth case, there is no data to be read in the buffer X or Y. At that time, the process returns to the beginning with the branch of arrow 516 or 517.

FIG. 9 shows how data is replenished by the disk read while delivering the data from the two buffers to the terminal. The shaded area in the figure indicates that data is included. As mentioned above, when one buffer is exhausted, the other buffer will deliver and data will be replenished during that time. The disk scheduler unit manages the bit rates of all video data, and the read request issuing unit knows the bit rate of the data currently delivered by inquiring the disk scheduler unit. Based on this, the deadline time at which the replenishment must be completed is calculated, and the deadline time is also given when the disk read request is issued. Assuming that the buffer size in the file server is a and the bit rate of video data is r, when a read request from the terminal comes in according to the bit rate, the time taken to exhaust one buffer is a /
r. The time after a / r after the disk read request is issued becomes the deadline time.

The file scheduler can guarantee a fast response to the terminal by performing the disk read while keeping the deadline.

However, it is usually considered that the application on the terminal also buffers several bytes. Therefore, the read request to the file server has a locally increased bit rate during buffering. At that time,
If the disk read is completed just before the deadline described above, it is possible that the disk read may not be completed in time. Therefore, it is necessary to set a margin at the deadline time and complete the disk read with a margin. The delivery processing unit sets the time after αa / r (where 0 ≦ α ≦ 1) as a deadline. by this,
Even if a request comes in at a temporarily high bit rate, delivery from the buffer can be performed.

The method of obtaining α is as follows. If the terminal side has a u-byte buffer and the file server side has an a-byte buffer, there will be at most u-byte fluctuations while the server is delivering data from the buffer. , A / u, there is a possibility that the deadline will be earlier. Therefore, α = a / u may be set.

There is a possibility that the terminal may request the file server not to reproduce the video data but to simply retrieve it from the file server. If it is simply taken out, the request may be made at a bit rate higher than the bit rate of the video data. In that case, if the delivery is carried out in accordance with a request for a high bit rate, the file server may put the CPU capacity to only one terminal, and the delivery to other terminals may be neglected.

Therefore, in the file server, the disk read is always performed after a certain period of time. In FIG. 9, the A buffer is replenished immediately after it is exhausted, whereas as shown in FIG. 10, even if A is exhausted quickly due to a request at a high bit rate from the terminal, the read request issuing unit does not The disk read request of A is issued to the scheduler only after αa / r from the time of requesting the disk read of A. At this time, there is a possibility that B will be exhausted before A is satisfied, and a state in which delivery cannot be performed immediately in response to a request from the terminal occurs. In this case, after the scheduler completes the disk read of A, the data delivery from the buffer of A can be resumed. By this processing, it is possible to avoid spending CPU power with only one delivery.

Next, a disk scheduling unit that receives a plurality of delivery processing units will be described. 12 and 1
As shown in FIG. 3, the disk read request issued from the delivery processing unit is accumulated in a queue in the disk scheduling unit.

There are three reasons for the delivery processing unit to issue a disk read request to the scheduler. One is disk read, which is necessary because the terminal has exhausted the data in the buffer unit as a result of continuously reading one data according to the bit rate of the data. Second, as a result of requesting data for special playback such as fast-forward at the terminal, the data in the buffer section becomes unnecessary,
This is a disk read that occurs because new data is needed. The third is the first disk read when a new terminal starts a read request. Modes are set for the above three disk reads, and mode 0, mode 1, and
Set to mode 2.

In mode 0, that is, in the case of normal continuous disk read, there is a deadline time in disk read. When the buffer size is a and the bit rate is r, the deadline time is a / r seconds after the disk read request is received. On the other hand, mode 1
In the case of mode 2 or, there is no deadline time.

As shown in FIG. 12, when the delivery processing unit issues a read request, the read buffer identifier, the file identifier, the read data offset, the size, the read deadline time, and the mode described above. give.

FIG. 13 is a block diagram of the request queue inserting section. The buffer identifier to be read is b, the file identifier is f, the offset of the data to be read is e, the size is s, the deadline time of reading is d, and the mode described above is p. Such information is stored in the queue. The algorithm for inserting in the queue is shown in FIG. In this algorithm, regarding the read request of mode 0, the deadlines are inserted so that they are always sorted in ascending order. When inserted, the scheduled read time t is calculated. This is determined by adding the disk read time performed by the request to the scheduled read time of the immediately preceding request.

If the scheduled read time is after the deadline, the deadline for this request will not be met. However, if there is a read request for data other than video data or a request for mode 2 or mode 1 in the requests that are ordered before the request, a new request can be generated by sending these requests later. Move forward. As a result, the deadline of the disk read can be satisfied or the delay with respect to the deadline can be minimized.

FIG. 11 shows a state in which a plurality of delivery processing units sequentially perform disk reading at fixed time intervals during a certain period T.

It is assumed that delivery is simultaneously performed to n terminals in a certain time period T. The terminal request distribution device knows the bit rate of the video data requested by the currently connected terminal and sets it as r1 to rn. 901-912
Z represents the time required for one disk read, and if the buffer size is a, the disk read interval of the terminal ri is a / ri, and considering the disk read interval with a margin, αa / ri (however, , 0 ≦ α ≦ 1)
Therefore, the number of disk reads within the time T is T / (αa /
ri) Therefore, the disk read time for all terminals is zT / (αa / r1) + zT / (αa / r2) + ...
+ ZT / (αa / r1) = zT (r1 + ... + rn) / αa
Becomes In order to be able to read all discs, this must be less than T, and zT (r1 + ... +
rn) / αa <T, that is, z (r1 + ... + rn) / α
The inequality a <1 holds. If this inequality holds when a new terminal is added, the terminal request distribution device permits the connection of the terminal. However, if it does not hold, the delivery to the terminal is refused. By this processing, it is possible to increase the number of terminals to the limit of the disk read capability within a range in which the terminal request can be delivered with high response.

[0040]

According to the file server of the present invention, it becomes possible to send the stored video data to more terminals with a quick response, and as a result, even if the number of terminals connected increases, Enables high quality video playback.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a block diagram of a terminal corresponding buffer processing unit.

FIG. 3 is an explanatory diagram of a file server of the present invention.

FIG. 4 is an explanatory diagram of a network environment to which the present invention is applied.

FIG. 5: After a data read request comes from the user terminal,
6 is a flowchart showing an algorithm in which a delivery processing unit delivers data to a user terminal.

6 is an explanatory diagram showing a state of a buffer in FIG.

FIG. 7 is an explanatory diagram showing a state of a buffer in FIG.

FIG. 8 is an explanatory diagram showing a state of a buffer in FIG.

FIG. 9 is an explanatory diagram showing how the delivery processing unit replenishes the buffer unit by disk read while delivering data from the two buffer units to the terminal.

FIG. 10 is an explanatory diagram showing the replenishment of the buffer unit when the terminal requests at a high bit rate for the purpose of copying in FIG.

FIG. 11 is an explanatory diagram showing a state in which a plurality of delivery processing units sequentially perform disk reading at fixed time intervals during a certain cycle T.

FIG. 12 is a block diagram of a disk scheduler unit according to the present invention.

FIG. 13 is a block diagram of a request queue insertion unit of the present invention.

FIG. 14 is a flowchart showing an algorithm in which a request queue inserting unit inserts a new disk read request into a queue.

[Explanation of symbols]

101 ... video file server, 111 ... terminal-compatible buffer processing unit, 121 ... network interface, 1
31 ... Terminal command distribution unit, 141 ... Disk scheduler unit, 151 ... Magnetic disk device.

Front Page Continuation (72) Inventor Yoshihiro Takiyasu 1099 Ozenji, Aso-ku, Kawasaki-shi, Kanagawa Hitachi, Ltd. Information & Communication Development Division

Claims (4)

[Claims] 1. A file server which has a storage unit for storing digital video data, and which delivers the digital data to the terminal in response to a read request for the digital data in the storage unit sent from a user terminal, temporarily stores the data. A buffer unit that has a faster read speed than the storage unit described above, and a request determination unit that determines whether the first read request sent from the user terminal is a video data request. If the data is the video data, the data reading unit that extracts more digital data from the first storage unit than the requested amount and stores it in the buffer unit, and the amount of the buffer unit that is requested by the terminal from the buffer unit A file server having means for taking out and sending out to a user terminal. 2. A means for determining a time (hereinafter referred to as a deadline time) at which the data in the buffer section underflows from the remaining data amount in the buffer section and the transmission bit rate according to claim 1, and before the deadline time is reached. A file server having a data reading unit that controls to read a predetermined amount from the storage unit. 3. Depending on whether or not the data requested by the terminal exists in the buffer unit according to claim 1, the data reading unit next replenishes the buffer unit by performing the current delivery in the buffer unit. Whether the request is a periodic read request for replenishing data that is continuous with the data in the existing portion or an aperiodic read request for replenishing data when the data requested by the terminal does not exist in the buffer section. A read request to the storage unit issued from the plurality of data reading units corresponding to the terminal, the deadline time thereof, and the distinction between aperiodic read and periodic read, and read to the storage unit. Create a queue of requests, sort the queue so that the deadlines are in ascending order with respect to periodic read requests, and finish reading all read requests in the queue. Time is calculated, and if there are things that the scheduled end time is later than the deadline time,
A file server having means for carrying back an aperiodic read request registered earlier in the queue to the back. 4. When storing data in the storage unit according to claim 1, whether or not the data is video data is registered together with the data, and the registered contents are referred to when reading the data. A file server including a request determination unit that determines whether the data to be read is video data.