CN101043445B - IO dispatch method in network storage system - Google Patents

Abstract

The invention provides an IO scheduling method in a network storage system. It includes the following computer-realizable steps: 1. A new IO command will enter the sending queue, and judge whether the command is an IO read and write response command or a short command belonging to a data flow control command; 2. If the command is an IO Read and write response commands or short commands belonging to data flow control commands, then go to step 3, if not, then go to step 4; 3, add the command to the head of the queue, return to step 1; 4, add the command to At the end of the queue, return to step 1. The present invention transmits the whole IO path or only the read-write response command in the IO command transmitted in the switch or the short command belonging to the data flow control, so as to improve the performance of the whole system and the resource utilization rate.

Description

IO Scheduling Method in Network Storage System

(1) Technical field

The invention relates to the field of network storage, in particular to the field of storage area network (Storage Area Network-SAN) in network storage technology.

(2) Background technology

Network storage technology uses a mature network interconnection instead of an IO bus to connect hosts and storage devices. In network connections, switches are often used. Fiber optic switches are used for fiber optic networks, and ordinary Ethernet switches are generally used for Ethernet, i.e. iSCSI storage systems. Switches currently used in network storage systems are generally not specially optimized for network storage applications. For example, for an iSCSI storage system, common Ethernet switches are generally used, and the switch is unknown to upper-layer applications, so it is impossible to optimize for network storage. In the Journal of Electronics, "A Preemptive Short Packet Priority Scheduling Algorithm in Input Queuing" in the fourth issue of 2005 proposed a short packet priority scheduling optimization algorithm for ordinary routers, but not an optimization method for network storage applications.

(3) Contents of the invention:

The purpose of the present invention is to provide an IO scheduling method in a network storage system that improves performance and resource utilization of the network storage system.

The object of the present invention is achieved like this: it comprises the steps that following several computers can realize:

1) A new IO command will enter the sending queue, and judge whether the command is an IO read and write response command or a short command belonging to a data flow control command;

2) If the command is an IO read/write response command or a short command belonging to a data flow control command, then go to step 3), if not, then go to step 4);

3) Add the command to the head of the queue and return to step 1);

4) Add the command to the end of the queue and return to step 1).

The present invention is characterized in that it is an optimization method for network storage applications, which prioritizes the entire IO path or only the read and write response commands in the IO commands transmitted in the switch or the short commands belonging to the data flow control command send.

The entire IO path in network storage technology includes hosts, storage networks and storage devices. In the read and write requests, the storage device waits for the read and write request commands from the host. If these request commands can be transmitted preferentially, the waiting time of the storage device can be reduced and the utilization rate of the storage device can be improved. After the system has completed the required operations, it usually waits for a response command before releasing the resources applied for the completion of the read and write requests. If these response commands can be transmitted in priority, the system can release resources as soon as possible and provide resources for subsequent read and write requests, thereby improving resource utilization. At the same time, because these commands are small packets that do not contain stored data, sending them preferentially can improve the system throughput.

The simulation test shows that improving the scheduling priority of the response command after reading and writing can significantly improve the resource utilization and performance of the storage system on the entire IO path. The present invention proposes that for network storage applications, the read and write response commands or short commands belonging to data flow control among the IO commands transmitted in the entire IO path or switch are transmitted preferentially, so as to improve the performance of the storage system and the utilization rate of resources.

(4) Description of drawings

Fig. 1 is the iSCSI read command flowchart in the embodiment of the present invention;

Fig. 2 is the iSCSI write command flowchart in the embodiment of the present invention;

FIG. 3 is a schematic diagram of an IO path in a network storage system.

(5) Specific implementation methods:

The present invention will be further described below in conjunction with accompanying drawing and specific embodiment:

In the network storage system, taking the iSCSI storage system as an example, its read and write process is as follows:

READ command:

When an initiator user program issues a read request to a target device:

①Convert the request into a SCSI command and pass it to the iSCSI underlying driver (Low Level Driver, LLD). iSCSILLD sends iSCSI "SCSI Command" PDU to the target after receiving the command.

②When the front end target driver (Front End Target Driver, FETD) of the target end receives this command, it decapsulates it and restores it to a SCSI command.

③ Pass this SCSI command to the SCSI Target Middle Level (SCSI Target Middle Level, STML).

④ STML returns the read data in the buffer to the initiator, that is, iSCSI "SCSI Data In" PDUs. After receiving the read data, the iSCSI LLD at the initiator saves it in the allocated empty buffer. After all data transmission is over, STML sends the response back to the initiator, that is, iSCSI "SCSI Respond" PDU.

⑤ After receiving the response, iSCSI LLD hands it over to SCSI Middle Level (SML) for processing. iSCSI LLD and STML release all resources allocated for this command.

Combined with Figure 1, the WRITE command:

When an initiator user program issues a write request to a target device:

① The request is converted into a SCSI command and passed to iSCSI LLD. iSCSI LLD sends iSCSI "SCSI Command" PDU to the target.

②When the target FETD receives this command, it decapsulates it and restores it to a SCSI command.

③ Pass this SCSI command to STML.

④ STML allocates the required buffer, sends iSCSI "Ready to Transfer" PDU to the initiator, and notifies the initiator that it can start sending data.

⑤ Once the iSCSI LLD receives the PDU, it sends the write data already stored in the buffer to the target, i.e. iSCSI "SCSI Data Out" PDUs.

⑥ FETD receives the write data from the start-up terminal and saves it in the allocated empty buffer. Then, notify STML that the write data has been received.

⑦ After all data transmission is completed, STML returns the processed response to the initiator, that is, iSCSI "SCSI Response" PDU.

⑧ After receiving the response, iSCSI LLD hands it over to SCSI Middle Level (SML) for processing. iSCSILLD and STML release all resources allocated for this command.

Combined with Figure 2, it can be seen from the reading and writing process of the iSCSI protocol that there are two types of iSCSI PDUs transmitted between the initiator and the target: one is the command containing read and write data, that is, iSCSI "SCSI Data In" PDUs and iSCSI "SCSI Data Out" PDUs; the other is commands related to resource waiting, such as request command iSCSI "SCSI Command" PDU, response command iSCSI "Ready to Transfer" PDU and iSCSI "SCSI Response" PDU. If these short commands are transmitted preferentially, the waiting time of resources can be reduced and resource utilization can be improved. For example, after reading and writing, the target end must send the iSCSI "SCSI Response" PDU command to the initiator end, and only after the initiator end receives this command, the initiator end and the target end can release all resources allocated for the read and write operation .

For each read and write operation, there will be many interaction processes between the initiator and the target end, and in the process of these information transmissions, most of the transmissions in the IO channel are requests for reading and writing and read and write completion commands or data flow control commands. short order. It has been proved by literature that if the scheduling priority of short packets is increased in the router, the performance can be improved. The present invention proposes to transmit the entire IO path or only the short commands in the IO commands transmitted in the switch first, so as to improve performance and resource utilization.

Fig. 3 is a schematic diagram of an IO path in a network storage system, the IO path includes a network sending device 1 for IO commands at the initiator side, a device 2 for processing IO commands in the switch, and a device 3 for network sending IO commands at the target side, Each of these IO commands are queued in a wait queue before being processed by them. 4, 5, and 6 are the waiting queues in the initiator, switch, and target, respectively.

In conjunction with FIG. 3 , the processing processes carried out in the network sending device 1 of the IO command at the initiator side, the device 2 for processing the IO command in the switch, or the device 3 of the network sending of the IO command at the target side are all the same, that is, the loops are respectively sent from the sending The head of the queue 4, 5 or 6 takes out the IO command and sends it to the network. The specific process is as follows:

(1) Take out the command in the sending queue.

(2) Send the command to the correct network.

(3) to step (1).

The initiator, switch, or target process the sending queues 4, 5, and 6 respectively as follows:

(1) A new IO command will enter the sending queue.

(2) Determine whether the command is an IO read and write response command, such as request command iSCSI "SCSI Command" PDU, response command iSCSI "Ready to Transfer" PDU and iSCSI "SCSI Response" PDU, if yes, go to step (3) , if not, go to step (4).

(3) Add the command to the head of the queue, go to step (1).

(4) Add the command to the tail of the queue, go to step (1).

Claims (1)

1. An IO scheduling method in a network storage system, characterized in that: the entire IO path or only the read-write response command in the IO command transmitted in the switch or the short order belonging to the data flow control command are sent preferentially, including the following Several computer-implementable steps: 1), a new IO command will enter the sending queue, and judge whether the command is an IO read and write response command or a short command belonging to a data flow control command; 2), if the command is an IO read/write response command or a short command belonging to a data flow control command, then go to step 3), if not, then go to step 4); 3), add the command to the head of the queue, and return to step 1); 4), add the command to the end of the queue, and return to step 1).

Images