Apparatus and method for performing cyclic redundancy check (CRC) on partial protocol data units (PDUS)

Abstract

The present invention discloses an apparatus and method for performing cyclic redundancy check (CRC) on partial protocol data units (PDUs). The disclosed apparatus is designed to off-load the CRC calculation for transmit or receive from a host computer. According to the disclosed method, when generating CRC for partial PDUs, for each such PDUs a decision is made to determine whether a CRC action is required, i.e., if CRC should be calculated, checked or placed in the outgoing byte stream. When partial CRC calculation is performed the intermediate value is saved into memory and later is used for calculating the CRC for a consecutive partial PDU. In accordance with a preferred embodiment of the invention, the need to re-calculate the CRC in a case of a re-transmit request is eliminated.

Description

TECHNICAL FIELD

The present invention relates generally to performing of cyclic redundancy check (CRC) on iSCSI protocol data units (PDUs), and more particularly for an apparatus and method for performing CRC on partial iSCSI PDUs.

BACKGROUND OF THE INVENTION

The rapid growth in data intensive applications continues to fuel the demand for raw data storage capacity. To meet this growing demand, the concept of the network storage systems was introduced. A network storage system is a network having a primary purpose of transferring of data between distributed computer systems and storage devices.

Network storage systems utilize the Internet Small Computer System Interface (iSCSI) protocol, which provides reliable data storage transport over a conventional transmission control protocol/Internet protocol (TCP/IP) network. The iSCSI protocol itself encapsulates small computer system interface (SCSI) commands in protocol data units (PDUs) carried in TCP/IP byte streams. That is, the iSCSI protocol allows network devices that are not connected by the same SCSI bus to communicate with each other over the Internet.

Data integrity is achieved by means of cyclic redundancy check (CRC) techniques. The CRC technique is used for checking and detecting errors in data transmitted over a network. The CRC algorithm and its underlying mathematics are well known to those skilled in the art. CRC generation is performed when data is transmitted from a host computer to the network. The CRC is calculated independently for a header and payload data portions included in an iSCSI PDU. The CRC is calculated for each portion independently and inserted to the PDU at locations reserved the CRCs calculated for the header and payload portions. CRC check is performed when an iSCSI PDU is received at the host computer. Here, the CRC is calculated and compared with a CRC value included in the PDU. The check is performed on both the header and payload portions.

Prior art implementations require the reception of an entire PDU before handling the CRC. Typically, an iSCSI PDU is composed of multiple data TCP segments that may have variable length size and further include data from more than one PDU. These segments are received in no particular order and multiple segments may be received from multiple different connections simultaneously. Therefore, in order to calculate the CRC related to the payload data of a PDU, prior art implementations construct the entire PDU before handling the CRC. Specifically, these implementations are not design to calculate the CRC of partial PDUs (e.g., TCP segments) while these partial PDUs are received or transmitted to the network. As an example, U.S. patent application Ser. No. 10/456,871 discloses a transport off-load engine (TOE) that performs CRC operations on iSCSI PDUs. The disclosed TOE receives a complete iSCSI PDU assembled by the host computer, calculates the CRC, and sends the result back to the host computer.

It would be therefore advantageous to provide an efficient solution for performing CRC operations on partial PDUs.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1—is a non-limiting exemplary block diagram of the apparatus for performing CRC operations on partial PDUs according to the present invention

FIG. 2—is an exemplary schema of an outbound byte stream

FIG. 3—is an exemplary layout of MODs array uses for forming a continuous outbound byte stream

FIG. 4—is a non-limiting flowchart describing the method for generating CRC in accordance with an exemplary embodiment of this invention

FIG. 5—is a non-limiting flowchart describing the method for checking CRC in accordance with an exemplary embodiment of this invention

FIG. 6—is a non-limiting diagram illustrates the entry states and theirs respective transitions

DETAILED DESCRIPTION

Referring to FIG. 1, there is shown a non-limiting exemplary block diagram of an apparatus 100 for performing CRC operations on partial PDUs, such as those that are received in conjunction with an Internet Small Computer System Interface (iSCSI). Apparatus 110 is designed to off-load the CRC calculation for transmit or receive from a host computer. Apparatus 100 includes a queue manager and scheduler (QMS) 110, a plurality of processing nodes (PNs) 120, a direct memory access (DMA) 130, a CRC controller 160, a transmit handler (TH) 170, a first memory 140 coupled to the DMA 130, CRC controller 160, and TH 170, and a second memory 150 coupled to the CRC controller 160. Typically, the first memory 140 is implemented using dynamic random access memory (DRAM), and the second memory 150 is implemented using static random access memory (SRAM). QMS 110 manages a plurality of queues, where each queue may have a plurality of memory object descriptors (MODs). The queues and MODs are located in the second memory 150, and the MODs are added to the queue as new host events, e.g., reception or transmission of a data segment. The MOD according to the present invention has pointers, including, but not limited to, a pointer to a memory location, such as a memory location in first memory 140, in host memory 190, or to a CRC placeholder in the second memory 150. If the MOD is the last MOD in the series, then the next MOD pointer may be set to null. A detailed description of the MODs is found in U.S. Pat. Nos. 6,760,304 (hereinafter the “304 patent”) and 10/219,673 (hereinafter the “673 application”) both by Oran Uzrad-Nali et al., assigned to common assignee, and which is hereby incorporated by reference for all that they disclosed. TH 170 calculates the CRC values of each segment as segments flow to the network. After processing the segments, the TH 170 routes the data in its network layer format to its destination.

Traffic is transmitted either in an outbound path or an inbound path. In the outbound path the CRC is generated as data is transferred from either first memory 140 or host memory 190 to the network. The CRC generation is performed by CRC controller 160 and TH 170 under the control of PNs 120. Once a PN 120 decides to move data it sets CRC controller 160 for the data movement and a CRC session is then established with TH 170 preparing it for the data movement. Subsequently, the respective PN 120 sends a series of commands to TH 170 to transfer the data and generate the CRC. At the end of the session, the respective PN 120 sends one or more messages to TH 170 to complete the process.

In the inbound path the CRC is checked as data, received from the network, is transferred from the first memory 140 to host memory 190. The CRC checking is performed by CRC controller 160 and DMA 130 under the control of PNs 120. Once a PN 120 decides to move data it sets the CRC controller 160 for the data movement and a CRC session is then established with DMA 130 preparing it for the data movement. Subsequently, the respective PN 120 sends a series of commands to DMA 130 to transfer the data and perform the CRC checking. At the end of the session, the respective PN 120 sends one or more messages to DMA 130 to complete the process.

In order to support CRC operation on partial PDUs, partial CRC values are maintained in second memory 150. In addition, CRC controller 160 includes a database of partial CRC values (hereinafter the “CRC-DB”) for maintaining partial CRC results. An entry in the CRC-DB includes the following fields: a state of the entry, a partial CRC value, and a location in second memory 150 of a partial CRC value. The entry's state may be one of: an idle, an idle error, a load requested, a load completed, an active, and a write back. A non-limiting diagram illustrates the entry states and theirs respective transitions is provided in FIG. 6. An idle state indicates that the entry is ready to be used in a next CRC session. An idle error state indicates that in a previous CRC session the entry used for a CRC check and a CRC error was detected. PN 120 may change the entry's state from an idle state or an idle error state only to the load requested state. The load requested state indicates that CRC controller 160 is retrieving a partial CRC value from second memory 150. As the partial CRC is fetched from second memory 150, the entry's state is changed to the load completed state. At the load completed state, the partial CRC value (or an initial CRC value) is available to be used by TH 170 or DMA 130. The entry's state is changed to the active state when the CRC value in fetched by the TH 170 or DMA 130. The active state indicates that a CRC session with TH 170 or DMA 130 is in progress. The entry's state is changed from the active state to the write back state, when the CRC session ends and TH 170 or DMA 130 requests to write back a new partial CRC. If the CRC check results with an error, the entry's state is changed to the idle error state. In the write back state, CRC controller 160 writes the partial CRC back to second memory 150. A partial CRC is the intermediate result of a CRC value calculated for a partial PDU.

Outbound byte stream that requires CRC generation comprises PDUs payload data reside in host memory 190 or first Memory 140, PDU headers reside in first memory 140. The outbound byte stream may be segmented to TCP segments. An example, FIG. 2 provides an exemplary outbound byte stream that includes three PDUs 210-1, 210-2, and 210-3, that are segmented into five segments 220-1 through 220-5. Each of PDUs 210-1 and 210-3 contain a header, payload data, and a CRC trailer for holding the generated CRC code. PDU 210-2 includes only a header portion. Each data block, i.e., a payload data in host memory 190 or a header in first memory 140 is pointed to by a MOD, and the MODs are linked together to create a continuous byte stream. The linkage of MODs is performed in one of the queues, e.g., a transmit queue (TTQ) managed by QMS 110 under the control of a PN 120 prior to transmitting the byte stream to the network. The process for adding MODs and linking them for the purpose of forming a continuous byte stream is described in greater detail in the '304 patent and the '673 patent application.

To determine whether a CRC code should be generated for a data block pointed to by a MOD, an indication is given to each MOD once it is added to its respective queue. Specifically, the MOD is colored to indicate whether the data block pointed to by the MOD is a payload data that is part of the CRC, the data block pointed by the MOD does not require CRC generation (e.g., PDU headers), or the MOD points to a CRC trailer in which the result of the CRC is inserted. In an exemplary embodiment of the present invention, MODs colored in green point to PDU payload and hence CRC generation is applied, MODs colored in blue point to CRC trailers, and MODs colored in red point to PDU headers and hence CRC action is not performed.

FIG. 3 provides a schematic diagram showing a layout of MODs pointing to PDU 210-1, 210-2, and 210-3 forming a continuous byte stream. The payload data of PDUs 210-1 and 210-3 are saved in host memory 190, while the header of PDUs 210-1, 210-2 and 210-3 are saved in first memory 140. TTQ 310 includes eight MODs 320-1 through 320-8, each may be colored either in green, blue, or red. MODs 320-1, 320-5 and 320-6 point to the headers of PDU 210-1, 210-2 and 210-3 respectively, and therefore are colored in red. MODs 320-2 and 320-3 point to two different locations in host memory 190 containing the payload data of PDU 210-1. These MODs are green, as CRC generation should be performed when processing the payload data pointed by them. MODs 320-4 and 320-8 are immediate MODs used as a placeholder for the CRC result of PDUs 210-1 and 210-3 respectively. Therefore, these MODs are colored in blue. MOD 320-7 is a green MOD that points to a location in host memory 190 containing the payload of PDU 210-3. A detailed example for generating CRC for the outbound stream depicted in FIG. 2 and FIG. 3 is provided below.

Referring to FIG. 4, a non-limiting flowchart 400 describing the method for generating CRC for partial PDUs in accordance with an exemplary embodiment of the present invention is shown. The CRC generation is performed through a CRC session established between CRC controller 160 and TH 170. At step S410, an entry is allocated in the CRC-DB. The state of the allocated entry is either an idle state or an idle error state. At step S420, the partial CRC field in the allocated entry is initialized with an initial CRC value. Alternatively, the partial CRC field may be set to with a partial CRC value retrieved from second memory 150. As a result, the entry state is set either to load requested or load completed. At step S430, a CRC session is established with TH 170. This is performed by instructing TH 170 to retrieve the initial CRC value from the allocated entry. TH 170 maintains a plurality of CRC channels, each of which includes CRC channel identification (ID), a pointer to an entry in the CRC-DB, and an intermediate CRC value. Once a CRC session is established, the pointer in the CRC channel points to the allocated entry and the initial value retrieved from the entry is saved in CRC channel. The implementation of a plurality of CRC channels allows for interaction of multiple PNs 120 with TH 170, and pipelining of CRC for better throughput.

At step S440, a ‘generate’ command is sent to TH 170 by a PN 120 that controls the process. The ‘generate’ command refers to a single data segment, of the byte stream, and includes a reference to the CRC channel. At step S450, a check is made to determine if it is required to calculate the CRC for the data segment that the command refers to. For this purpose, TH 170 monitors the color of the MOD associated with the respective data segment, as the segment flows to its target network interface. At step S453, a check is made to determine as whether the MOD is coded in red. If so, the execution continues with step S450 where the next MOD is processed; otherwise, execution continues with step S455. At step S455, the color of the MOD is determined, and if the MOD is coded in green, i.e., CRC calculation is required, then at step S460 a partial CRC for the segment is calculated. If the MOD is coded in blue, i.e., CRC insertion is required then at step S470 the current CRC result inserted into the outgoing byte stream at the CRC trailer. The execution continues with step S472 where the CRC value of the CRC channel is set to the initial value, enabling the CRC calculation of the next PDU in the byte stream. At step S474, the CRC result is written to a placeholder MOD in second memory 150, in order to avoid the need for re-calculating the CRC upon TCP re-transmit request. If a re-transmit request was sent, then TH 170 identifies the MOD retrieves the CRC result from second memory 150 and sends it back as part of the byte stream. The CRC result is removed from second memory 150 when the TCP acknowledged is received.

At step S480 the CRC session with TH 170 is closed and the CRC channel writes back the current channel CRC value to the allocated entry. As a result, the entry's state is changed to a write back state, and the intermediate CRC value is saved to second memory 150 to a location designed by the partial CRC location field.

It should be noted by one who is skilled in the art that the CRC calculation is performed as PDUs flow toward the network, hence, the CRC calculation is performed without consuming additional bandwidth from first memory 140.

Following is a detailed example for generating CRC for the outbound byte stream shown in FIG. 2 and the respective MODs layout shown in FIG. 3. The byte stream is transmitted to the network in two CRC sessions. At the first CRC session, only segments 220-1 and 220-2, with 512 bytes each are sent. Upon receiving an event from the QMS 110 an entry in the CRC-DB is allocated and subsequently this entry is initialized. The entry is initialized with an initial CRC value, since segment 220-1 is the beginning of PDU 210-1. Next, a CRC session with TH 170 is established while referring a CRC channel to the allocated entry and thereafter two ‘generate’ commands for segment 220-1 and segment 220-2 are sent to TH 170 by PN 120. For segments 220-1 and 220-2 only MODs 320-1 through 320-3 are monitored. MOD 320-1 is coded ‘red’ and therefore a CRC action is not performed, MODs 320-2 and 320-3 are coded in green and therefore the CRC for the payload data pointed by these MODs is calculated. After processing segment 220-2, the first CRC session is closed and the partial CRC is written back to the allocated entry and to a memory location in second memory 150 associated with this connection.

A CRC second session is established when a new event to transmit segments 220-3 and 220-4 is received from QMS 110. It should be further noted that events for other connections may arrive between the end of the first session and the beginning of the second session. Segment 220-3 is not at the beginning of a PDU, and thus TH 170 allocates an entry in the CRC-DB and retrieves the partial CRC associated with the connection. The memory location of the partial CRC is stored in the connection context. PN 120 issues a CRC ‘generate’ command with a reference to the CRC channel for both segments 220-3 and 220-4. For segments 220-3 and 220-4 only MODs 320-3 through 320-7 are monitored. MOD 320-4 is coded in blue and therefore TH 170 replaces the CRC trailer with the CRC results, initializes the CRC channel in order to prepare it for PDU 210-3 and requesting CRC controller 160 to write the CRC result to a location in second memory 150.

Referring to FIG. 5, a non-limiting flowchart 500 describing the method for checking CRC for partial PDUs in accordance with an exemplary embodiment of the present invention is shown. The CRC check is performed through a CRC session established between CRC controller 160 and DMA 130. At step S510, an entry is allocated in the CRC-DB. The state of the allocated entry is either an idle state or an idle error state. At step S520, the partial CRC field in the allocated entry is initialized with an initial CRC value. Alternatively, the partial CRC field may be initialized with a partial CRC value retrieved from second memory 150. As a result, the entry state is set to a load requested or a load completed. At step S530, a CRC session is established with DMA 130 by instructing it to retrieve the initial CRC value from the allocated entry. DMA 130 maintains a plurality of CRC channels, each of which includes CRC channel identification (ID), a pointer to an entry in the CRC-DB, and an intermediate CRC value. Once a CRC session is established, the pointer in the CRC channel points to the allocated entry and the initial value retrieved from the entry is saved in the CRC channel. The implementation of a plurality of CRC channels allows for interaction of multiple PNs 120 with DMA 130.

At step S540, a ‘check’ command is sent to DMA 130 by a PN 120 that controls the process. The ‘check’ command refers to a single data segment and includes a reference to the CRC channel. CRC check is performed on payload data after TCP processing and on delineated PDUs. PN 120 that controls the process is aware of the position of the segments within the PDU. For a segment placed at the middle or the beginning of a PDU, the CRC is calculated. For a segment at the end of a PDU, DMA 130 compares the calculated CRC result with the content of CRC trailer. At step S550, The comparison result is reported back to CRC controller 160 and if an error was detected then the state of the allocated entry is changed to an idle error state. At step S560, the CRC session with DMA 130 is closed. As a result, the CRC channel writes back the current channel CRC value to the allocated entry. As a result, the entry's state is changed to a write back state and the intermediate CRC result is saved to second memory 150 to a location designed by the partial CRC location field.

Other modifications and variations to the invention will be apparent to those skilled in the art from the foregoing disclosure and teachings. Thus, while only certain embodiments of the invention have been specifically described herein, it will be apparent that numerous modifications may be made thereto without departing from the spirit and scope of the invention.

Claims

1. An apparatus for performing CRC operations on protocol data units (PDUs), the apparatus comprising: a queue manager and scheduler capable of managing a plurality of queues, wherein each of queues comprise a plurality of memory object descriptors (MODs); a CRC controller capable of performing said CRC operations; a plurality of processing nodes capable of controlling said CRC operations; a transmit handler capable of calculating said CRC values of segments flow to a network; a first memory for maintaining at least said partial PDUs; and, a second memory for maintaining at least said plurality of MODs and partial CRC values; whereby said apparatus is designed to off-load the CRC calculation for PDUs transmitted or received from a host computer.

2. The apparatus of claim 1, further comprising: a direct memory access coupled to said CRC controller, said first memory and a host memory.

3. The apparatus of claim 1, wherein said CRC controller further includes a database of partial CRC values.

4. The apparatus of claim 1, wherein said CRC operations comprises at least one of: generating a CRC value, checking a CRC value, inserting a CRC value in a CRC trailer, re-transmitting a CRC value.

5. The apparatus of claim 1, wherein each of said MOD points to a data block.

6. The apparatus of claim 5, wherein each of said MOD is marked with a different color to indicate whether said data block pointed by said MOD requires a CRC generation.

7. The apparatus of claim 1, wherein said PDUs are of the fifth layer of the standard communication model.

8. The apparatus of claim 1, wherein said apparatus is further configured to generate a CRC for a partial PDU.

9. The apparatus of claim 8, wherein the generation of CRC for a partial PDU comprises the steps of: identifying the color of the MOD respective of said partial PDU; waiting for MOD color change if MOD color is red and then proceeding with the following steps; calculating the CRC if the MOD color is not blue; otherwise, performing the following steps: inserting CRC result in trailer; resetting the CRC channel; writing CRC result to memory.

10. The apparatus of claim 1, wherein said apparatus is further configured to check a CRC for a partial PDU.

11. The apparatus of claim 10, wherein the check of CRC of a partial PDU comprises the steps of: sending a check command per segment; and, reporting the CRC check results.

12. The apparatus of claim 1, wherein said apparatus performed the CRC generation upon detection of a partial iSCSI PDU.

13. A computer network configured with the apparatus for performing CRC operations on protocol data units (PDUs) of claim 1.

14. A method for generating a CRC for a partial PDU comprising the steps of: identifying the color of the MOD respective of said partial PDU; waiting for MOD color change if MOD color is red and then proceeding with the following steps; calculating the CRC if the MOD color is not blue; otherwise, performing the following steps: inserting CRC result in trailer; resetting the CRC channel; writing CRC result to memory.

15. The method of claim 14, wherein said partial PDU is a partial iSCSI PDU.

16. The method of claim 14, wherein said method is operative to off-load the CRC calculation for PDUs transmitted or received from a host computer.

17. A computer network configured to operate the method for generating a CRC for a partial PDU of claim 14.