JP5179218B2 - Method for controlling number of TCP connections of iSCSI session, iSCSI host device, and iSCSI initiator configuration program - Google Patents

Description

  The present invention relates to an iSCSI technology for connecting an iSCSI host device such as a file server and an iSCSI device device such as a storage via an IP network, and more particularly to a technology for improving throughput.

(1) iSCSI (Internet SCSI)
A storage device composed of a single or a collection of a plurality of hard disk drive devices, or a disk array storage device that controls a plurality of hard disk drives with a dedicated control unit has been put into practical use.

  Examples of interface technologies for connecting a storage device (device device) and a host device (server, PC) include Fiber Channel (FC), InfiniBand, SCSI (Small Computer Systems Interface), and the like.

  The SCSI interface is excellent as an application for economically connecting a short distance, and is already widely used. In the interface, a function corresponding to a client is called an initiator, and a function corresponding to a server is called a target, and a SCSI command (CDB: COMMAND Descriptor Block) is exchanged between the initiator and the target.

  As shown in Non-Patent Document 1, iSCSI (Internet SCSI) technology was standardized by the IETF (The Internet Engineering Task Force) as an upper protocol of TCP / IP as a network protocol in 2004 (RFC3720). Unlike other network storage protocols such as Fiber Channel, this iSCSI connection can be used if the computer has an Ethernet interface.

Currently, the iSCSI technology is used for a line quality of gigabit or higher, particularly in a backbone line such as in a data center, between data centers, and a large-scale LAN system. According to Non-Patent Document 1, the iSCSI technology defines a method of using a plurality of TCP connections, but it is not necessarily implemented because of an optional function.
(2) High-speed iSCSI technology On the other hand, the effectiveness of using multiple TCP connections is known as a method for preventing a decrease in iSCSI throughput due to network delay on a wide-area / wide-band IP network (Non-patent Document 2. 3). In addition, Non-Patent Document 4 that uses a multilink as a means for realizing multiple use of TCP connections, Non-Patent Document 5 that changes the protocol of the transport layer, and the like have also been proposed.

  For example, Non-Patent Document 4 relating to a technique using a multilink proposes a technique for establishing a parallel TCP connection through a plurality of routes and using a VPN multihoming function to improve throughput.

  Further, in Non-Patent Document 5 regarding a method of changing the protocol of the transport layer, a plurality of LAN ports are used to physically multiplex links and establish a TCP connection through different paths in order to improve throughput. Yes.

However, Non-Patent Document 4 has a limitation in the environment in which multilink can be used, and Non-Patent Document 5 has a problem that it is difficult to connect to a multi-vendor device.
RFC370 “Internet Small Computer Systems Interface-face (iSCSI)”, Apr 2004 Q. K. Yang, “On perforation of parallel iSCSI protocol-worked storage systems,” In Proceedings of the 20th International Conference on Advanced Invented. 1, pp. 629-636, Apr. 2006. G. Motwani and K.M. Gopinath, “Evaluation of advanced TCP stacks in the iSCSI envi- ronment using simulation model,” Proceedings of the 22nd IEEE / 13th NASA Standard. 210-217, 2005. B. K. Kancheria, G. M.M. Narayan, and K.K. Gopinath, “Performance evaluation of Multiple TCP connection in iSCSI,” in Proceedings of the 24th IEEE Conference on Mass Storage Systems and Technologies. 239-244, IEEE Computer Society, Sept. 2007 Nozomi Chishima, Minoru Yamaguchi, Masato Oguchi, “Performance and TCP parameter analysis during VPN multi-path access in iSCSI storage,” IEICE 18th Data Engineering Workshop DEWS2007, Feb. 2007. L. Qiu, Y .; Zhang, and S.M. Keshav, “On individual and aggregate TCP Performance,” in Processes of Internet Conference on Network Protocols, pp. 203-212, Oct. 1999. T.A. Ito, H.C. Ohsaki, and M.M. Imase, “On parameter tuning of data pour in the World of pour.” In Processes of Second International ps., “Non-parameter tuning of data ptr. 415-421, Oct. 2005. T.A. Ito, H.C. Ohsaki, and M.M. Imas, “GridFTP-APT: Automatic parallel tuning mechanism for data transfer protocol (GTP)” in Proceedings of 6th IEEE Criteria. 454-461, May. 2006.

  However, although the problem of a decrease in iSCSI throughput in a wide-area / broadband IP network is widely known, this problem cannot be solved effectively in the past.

  In other words, for iSCSI connections based on wide-area and wide-band IP networks, to improve throughput, parameters such as iSCSI / TCP are tuned in advance based on the bandwidth and delay information of the line used, or protocol conversion is performed between the server and storage. Operation settings are complicated, such as installing a machine to improve throughput, and there is a risk that economic operation will be difficult.

  In addition, because such improvements are based on fixed information, it is difficult to follow bandwidth fluctuations, making it difficult to transfer data efficiently, and there is a risk that a best-effort and economical line cannot be used. there were.

  Therefore, the present invention has the problem that the throughput of iSCSI decreases in a wide-area / broadband IP network. (1) While using an existing storage device, (2) limiting the use environment such as network delay and use bandwidth. (3) The problem is to solve economically.

  Although the iSCSI technology has a function of establishing a plurality of TCP connections in one iSCSI session, issuing a plurality of commands in parallel, and simultaneously transmitting and receiving data through each TCP connection, Does not describe how to determine the number of TCP connections.

  The present invention is a technical idea created by paying attention to the property that the throughput of a parallel TCP connection becomes a convex function with respect to multiplicity, and the multiplicity of TCP connection is optimized with respect to the throughput. Has solved.

  Specifically, the invention described in claim 1 is a method for controlling the number of TCP connections with an initiator function built in an iSCSI host device when the iSCSI host device and the iSCSI device device are connected by an iSCSI session. The number of TCP connections used to transmit / receive the fixed amount of data to be transmitted / received next with reference to the ratio of the fixed amount of data transmitted / received through the iSCSI session to the time required to transmit / receive the amount of data And a second step of increasing / decreasing the number of TCP connections used for the next transmission / reception based on the determination.

  According to a second aspect of the present invention, the first step includes a step of performing transmission / reception with the iSCSI device apparatus by distributing to a fixed amount of data for each TCP connection, and transmission / reception for each TCP connection transmitted from the iSCSI device apparatus side. Obtaining a time required for the transmission / reception based on a completion notification.

  According to a third aspect of the present invention, in the first step, the data transmission / reception throughput value is measured while increasing the number of TCP connections, and when the throughput value is lower than the previous value, the numerical range of the number of TCP connections is extracted. And a step of narrowing down the number of TCP connections by applying the golden search method to the numerical width and determining an optimum value of the number of TCP connections that maximizes the throughput value.

  The invention according to claim 4 is an iSCSI host device having an initiator function connected to an iSCSI device device through an iSCSI session, and transmits and receives a certain amount of data through the iSCSI session. Means for determining the number of TCP connections used for transmitting / receiving a certain amount of data to be transmitted / received next, and TCP connections used for transmitting / receiving next based on the determination It functions as a means for increasing / decreasing the number.

  According to a fifth aspect of the present invention, the transmission / reception is performed on the basis of a means for performing transmission / reception with the iSCSI device apparatus by distributing a fixed amount of data for each TCP connection, and a transmission / reception completion notification for each TCP connection transmitted from the iSCSI device apparatus side. It is characterized by functioning as a means for obtaining the time required for.

  According to a sixth aspect of the present invention, there is provided means for measuring a throughput value of data transmission / reception while increasing the number of TCP connections, and extracting a numerical width of the number of TCP connections when the throughput value is lower than the previous value; It is characterized by functioning as means for determining the optimum value of the number of TCP connections that maximizes the throughput value by narrowing down the number of TCP connections by applying the golden search method to the width.

  A seventh aspect of the invention is characterized in that a computer functions as the iSCSI host device according to the fourth to sixth aspects.

  According to the first to seventh aspects of the invention, the optimum value of the number of TCP connections is automatically calculated, and the number of TCP connections is operated with the value. Therefore, operation and installation can be simplified, and it is possible to follow bandwidth fluctuations. As a result, efficient data transfer can always be realized on a given line, and economic efficiency is improved.

  In particular, according to the seventh aspect of the present invention, it is sufficient to install a program in an existing iSCSI host device, and the economics of system construction / update are further improved.

  Hereinafter, embodiments of the present invention will be described with reference to FIGS. FIG. 1 shows a basic form in which an iSCSI initiator 1 and an iSCSI target 2 are iSCSI-connected via an IP (internet protocol) network.

  The iSCSI initiator 1 is mounted on a file server that configures an iSCSI host device, while the iSCSI target 2 is mounted on a disk storage device that configures an iSCSI device device. Here, a plurality of TCP connections can be formed in an iSCSI connection, and the number of TCP connections is expressed as “N”.

  The function of the iSCSI initiator 1 is realized and constructed through a communication device (for example, an Ethernet interface or the like) by reading a program code of an initiator configuration program by a processing unit (CPU “Central Processing Unit” or the like) of a file server.

  The file server is configured by a computer and includes a memory (RAM) that can temporarily store processing data, an input device such as a keyboard and a mouse, a display unit such as a monitor, and the like.

  FIG. 2 shows a configuration in which a control function for the number of TCP connections N (iSCSI-APT “Automatic Parallelism Tuning” 3 in FIG. 2) is added to the function of the iSCSI initiator 1. Similar to the function of the iSCSI initiator 1, this control function is implemented and constructed by the processing unit of the file server reading the program code of the initiator configuration program.

  Here, the file server mounted with the iSCSI initiator 1 divides the data written (transferred) into the storage device mounted with the iSCSI target 2 into data (chunks) having a substantially constant size even if the number of TCP connections N varies. Transfer while.

  At this time, the chunk is an integer multiple of the “iSCSI PDU” size. Therefore, if “n” pieces are distributed almost equally to each TCP connection with respect to the number of TC connections “N”, the following formula 1 is established.

Hereinafter, operation examples S1 to S7 will be described with reference to FIG.
(1) Operation example of S1 to S7 Processing of S1: First, transfer data requested to be written to the iSCSI layer through an upper layer in the file server in which the iSCSI initiator 1 is arranged, for example, a file system, is transferred to the TCP connection. The data is divided into uniformly sized data (chunks) independent of the number N. If the size of the chunk is too small, the throughput cannot be accurately reflected due to the characteristics of the TCP / IP protocol. Therefore, the size of the chunk is set to a size sufficient to calculate the throughput almost accurately.

  S2 processing: iSCSI-APT3, which is the processing function of the present invention, establishes an initial TCP connection and performs "iSCSI login" authentication in order to make an iSCSI connection to iSCSI target 2 (storage) (usually here) There is one TCP connection to be connected.)

  Processing in S3: The chunk generated in the processing in S1 is divided into “iSCSI PDU” size which is an iSCSI data transfer unit, and the divided “iSCSI PDU” is almost equal to “n” TCP connections. Sort out.

  Table 1 shows an example of the number N of TCP connections of “Table 1” and chunk division. Here, “chunk (n): iSCSI PDU number / TCP session” is taken as an example, and it is shown that the number of TCP connections “N” decreases to “1 / N” per TCP connection.

  Processing in S4: “iSCSI PDUs” (about n) to be transferred for each distributed TCP connection are transferred through each TCP connection. At this time, as shown in FIGS. 3 and 4, the iSCSI-APT 3 records the time when the transfer is first started among all the TCP connections.

  Process of S5: As shown in FIG. 3, the SCSI target 2 sequentially transmits “iSCSI PDU” reception completion notification, that is, R2T (Ready to transfer) for each TCP connection.

  Process of S6: iSCSI-APT3 records the time when R2T was last received in all TCP connections, and, as shown in FIG. 4, throughput G (N ) Is calculated. Immediately thereafter, the number N of TCP connections to be set for the next chunk transfer is determined based on a built-in algorithm process described later.

Processing in S7: In order to change to the number N of TCP connections set for the next chunk transfer, TCP connection increase / decrease control is performed.
(2) Built-in Algorithm Processing Here, an example of determining the number of TCP connections N performed in S6 will be described with reference to FIG. Here, there is shown a property that the throughput (chunk / transfer time T) of the parallel TCP connection is an upward convex function with respect to the multiplicity of the number N of connections.

  Table 2 shows “Table 2” used in the processing example of FIG. In “Table 2”, the initial value No. of the TCP connection number N is set to “1”, and the control parameter k is set to “2”.

The process of optimizing the number N of TCP connections consists of the following stage 1.2. The stage 1 corresponds to the transition of “A to E” shown in FIG. 5, and the stage 2 corresponds to the transition of “E to H” shown in FIG. 5.
(1) Stage 1
Stage 1 increases the number N of TCP connections from the initial value N ₀ by multiplying by the control parameter k, and provides a rough width (hereinafter referred to as a bracket) of the number of connections (horizontal axis in FIG. 5) including the optimum value Nopt. Explore. This state is shown in Equation (2).

  After that, based on the method described in the processing of S3 to S6, the chunk transfer throughput G (N) is calculated. Until the calculated throughput G (N) is in the relationship of the following formula (3) with the previous calculated value,

  Based on the following equation (4):

  While increasing the number N of TCP connections, the chunk throughput G (N) is measured, and the expression (3) is evaluated (compared with the previous value). When the expression (3) is satisfied (TRUE) (when the throughput is lower than the previous value), the first bracket has three N values in the history for the past three times as shown in the expression (5). Is set as a group, and the process proceeds to stage 2.

In FIG. 5, “A” is an initial value, and “N = N ₀ = 1” as described in “Table 2”. Next, the TCP connection is doubled (“B” in FIG. 5) using the parameter k described in “Table 2”. Further, in the next chunk transfer, the evaluation of Expression (3) becomes Expression (6) when N = 4 (“C” in FIG. 5) and N = 8 (“D” in FIG. 5) are increased. The first (1st) bracket (2, 4, 8) is obtained ("E" in FIG. 5).

(2) Stage 2
In stage 2, the golden search method is applied to the width of the first bracket (horizontal axis in FIG. 5) including the optimum value Nopt calculated in stage 1 (the value of N indicating the only maximum point of the throughput of the chunk). Narrow down and extract the optimum value Nopt. For this, the method of Non-Patent Document 8 is mainly used. In this non-patent document 8, the golden search method is used for another technique called GridFTP instead of iSCSI. Hereinafter, the narrowing of the stage 2 will be specifically described.

  First, “N”, which is a new number of TCP connections, is calculated from the first bracket extracted in stage 1 by using the golden ratio (Golden Rasio) of Equation (7) and Equation (8).

  For the bracket (l, m, r)}, Equation (9) is applied to calculate a new TCP connection number N (rounded up after the decimal point).

  Next, the chunk transfer throughput G (N) is calculated based on the method described in S3 to S6. When the relationship of Expression (7) is true (TRUE), a new bracket is introduced according to Expression (10).

  When the relationship of equation (7) does not hold (FALSE), a new bracket is derived from equation (11), and a new number N of TCP connections is calculated from equation (9) using the newly derived bracket. To do.

  When the above process is repeated and (l, m, r) becomes a continuous integer, Expression (12) is obtained.

  Actually, when the 1st bracket (2, 4, 8) extracted in stage 1 is applied to equation (9), equation (13) is obtained.

  When Expression (7) is evaluated with the result (vertical axis) in FIG. 5 when N = 6, Expression (14) is established.

Therefore, the 2nd bracket of the formula (15) is configured by the formula (11) (“E” in FIG. 5).

  When “N” is calculated again by equation (9), equation (16) is obtained.

  Formula (7) is evaluated at N = 7 (see FIG. 5), and the 3rd bracket of Formula (18) is configured by Formula (10) from the result of Formula (17) (“F” to “G” in FIG. 5). ).

  When N is calculated again by equation (9), equation (19) is obtained.

  Formula (7) is evaluated with N = 5 (see FIG. 5), and the 4rd bracket of Formula (21) is configured by Formula (11) from the result of Formula (20).

  Here, since all the integers in the bracket are continuous, stage 2 is ended. As the optimum value Nopt of the number of TCP connections that maximizes the throughput, “6” that is the number at the center of the last bracket shown in Expression (22) can be extracted.

In summary, in the example of FIG. 5, the transition shown in Expression (23) is performed until the initial value N _{0 of the} number of TCP connections is shifted to the optimum value Nopt.

  In this way, the TCP (through the chunk / transfer time T) of the parallel TCP connection becomes a convex function with respect to the multiplicity of the connection N (“A” to “H” in FIG. 5), and TCP The multiplicity of connections is optimized with respect to the throughput (the number of TCP connections that maximizes the throughput is searched and the increase / decrease control is performed up to the number of TCP connections).

  In other words, the iSCSI-APT3 control function that automatically adjusts the number N of TCP connections incorporated in the iSCSI initiator 1 allows the processing function of the iSCSI-APT3 to be performed periodically even when there is no information such as delay characteristics or bandwidth of the network to be used in advance. By operating automatically, the TCP connection number Nopt optimized for the network at the time of operation is automatically followed.

  As a result, the throughput between the file server connected to the iSCSI and the storage can be increased. As a result, the operation and installation of a file server or the like can be greatly simplified and the bandwidth fluctuation can be tracked, so that efficient data transfer can always be realized on a given line.

  Further, since it is sufficient to modify the iSCSI initiator in terms of the file server implementation, a commercially available target device that supports the TCP connection multiplexing option can be used, and in this sense, an economical system construction / update is possible.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation is possible in the range described in each claim. For example, the same effect can be expected not only for data transfer (transmission) but also for data reading (reception).

  The iSCSI initiator 1 incorporating the iSCSI-ATP3 control function can also be constructed as a program that causes a computer to function.

  In this case, the processing unit (for example, CPU) of the computer reads the program code, and the process of stage 1.2 is executed through the communication device.

  Therefore, the file server having the iSCSI initiator function according to the present embodiment can be constructed simply by installing the program in the file server in the existing iSCSI system. In this respect, the economics of system construction / update are further improved.

  This program code is stored in a recording medium such as a CD-ROM, DVD-ROM, CD-R, CD-RW, DVD-R, DVD-RW, MO, and HDD. The program may be downloaded from an Internet site and provided to a computer.

Explanation of symbols

1 ... iSCSI initiator 2 ... iSCSI target (iSCSI device)
3 ... iSCSI-SPT
N ... Number of TCP connections G (N) ... Throughput value

Schematic diagram of iSCSI connection using a plurality of TCP connections. FIG. 4 is a schematic diagram of iSCSI connection of an iSCSI initiator with a TCP connection number control function (iSCSI-APT) according to an embodiment of the present invention. The schematic diagram of the time definition for chunk transfer time calculation which shows the example of a sequence of "iSCSI PDU" transfer by one TCP connection. Schematic which shows the definition of the transfer time in the chunk transfer using a some TCP connection. Schematic which shows the process example of deriving the optimal number of TCP connections from observation of the throughput of chunk transfer.

Claims (7)

A method of controlling the number of TCP connections with an initiator function built in an iSCSI host device when the iSCSI host device and the iSCSI device device are connected by an iSCSI session,
The number of TCP connections used to transmit / receive the fixed amount of data to be transmitted / received next is determined by referring to the ratio of the fixed amount of data transmitted / received through the iSCSI session to the time required to transmit / receive the data amount. A first step to determine;
A second step of increasing / decreasing the number of TCP connections used for next transmission / reception based on the determination;
A method for controlling the number of TCP connections of an iSCSI session, comprising: The first step is a step of performing transmission / reception with the iSCSI device device by distributing to a certain amount of data for each TCP connection;
Obtaining a time required for the transmission / reception based on a transmission / reception completion notification for each TCP connection transmitted from the iSCSI device side;
The method of controlling the number of TCP connections of an iSCSI session according to claim 1, wherein: The first step is a step of measuring a throughput value of data transmission / reception while increasing the number of TCP connections, and extracting a numerical width of the number of TCP connections when the throughput value is lower than the previous value;
Applying the golden search method to the numerical width to narrow down the number of TCP connections, and determining an optimum value of the number of TCP connections that maximizes the throughput value;
The method for controlling the number of TCP connections of an iSCSI session according to claim 1, wherein: An iSCSI host device having an initiator function connected to an iSCSI device device through an iSCSI session,
The number of TCP connections used to transmit / receive the fixed amount of data to be transmitted / received next is determined by referring to the ratio of the fixed amount of data transmitted / received through the iSCSI session to the time required to transmit / receive the data amount. Means to determine,
Means for increasing / decreasing the number of TCP connections used for next transmission / reception based on the determination;
An iSCSI host device characterized by functioning as Means for performing transmission / reception with the iSCSI device by allocating a fixed amount of data for each TCP connection;
Means for obtaining a time required for the transmission / reception based on a transmission / reception completion notification for each TCP connection transmitted from the iSCSI device side;
The iSCSI host device according to claim 4, wherein the iSCSI host device functions. Means for measuring a throughput value of data transmission / reception while increasing the number of TCP connections, and extracting a numerical width of the number of TCP connections when the throughput value is lower than the previous value;
Means for narrowing down the number of TCP connections by applying the golden search method to the numerical range, and determining an optimum value of the number of TCP connections that maximizes the throughput value;
6. The iSCSI host device according to claim 4, wherein the iSCSI host device functions.   7. An iSCSI initiator configuration program for causing a computer to function as the iSCSI host device according to claim 4.

Images