|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] TCP Framing Support in iSCSI - ProposalThe following rough proposal for iSCSI TCP Framing Support has been generated by the Framing Team as a result of the Face-to-Face Design Teams meeting and subsequent discussions. It consists of changes to the "ULP Framing for TCP" and "iSCSI" internet-drafts as summarized below. An I-D detailing these, and other related, changes is forthcoming. Regards, Jim Wendt jim_wendt@hp.com -------- iSCSI TCP Framing Support Proposal - Summary ------ A) Proposed changes to the "ULP Framing for TCP" I-D are: 1) Modify I-D to include two framing modes: - "Marker mode" for unmodified TCP stacks - "PDU-alignment mode" for modified TCP stacks Note: An updated version of the "ULP Framing for TCP I-D" reflecting these changes has been posted (7/9/01) to TSVWG for comments (draft-ietf-tsvwg-tcp-ulp-frame-00) 2) ULP is responsible for negotiating use of a framing mode and enabling framing behavior on the TCP connection in an unambiguous manner. 3) There are "Senders" and "Receivers" on each unidirectional data flow of a framing TCP connection. The use of framing, the framing mode, and framing operational parameters values are negotiated separately for each direction of a framing TCP connection. This means that framing behavior may be in use on one direction of a TCP connection but not the reverse direction, or that different framing modes may be in use. 4) Creation of the following operational parameters and semantics: - Marker period (in Marker mode) - Receiver maximum PDU size (in Marker mode) - Framing keys (in PDU-alignment mode) - ULP packing behavior (in PDU-alignment mode) 5) ULP is responsible for negotiating use of a specific framing mode over the TCP connection, and for negotiating values for the framing operation parameters 6) Change the marker fields to be 16-bits rather than 32-bits (and refer to as "offsets" rather than "pointers") B) Proposed changes to the iSCSI spec are: 1) Remove Markers appendix from iSCSI spec (Appendix D. Synch and Steering with Fixed Interval Markers) 2) iSCSI spec adds wording to the effect of: Framing Protocol: * iSCSI initiator and target framing behavior over a TCP connection is defined in draft-ietf-tsvwg-tcp-ulp-frame-00 (or eventual RFC#) * an iSCSI initiator or target is both a sender and receiver with respect to framing behavior over a TCP connection. iSCSI Sender: * an iSCSI framing sender SHOULD implement PDU-alignment mode, as defined in <Framing I-D> * if an iSCSI framing sender does not implement PDU-alignment mode, it MUST implement Marker mode, as defined in <Framing I-D> iSCSI Receiver: * an iSCSI receiver may choose to not implement any framing mode * an iSCSI framing receiver MAY implement PDU-alignment mode, or Marker mode, or both as defined in <Framing I-D> * an iSCSI framing receiver on a TCP connection dictates use of the highest framing mode desired from the sender by progressing through the following sequence: - if the receiver requests PDU-alignment mode from the sender, and the sender supports PDU-alignment mode, then the sender MUST enable use of PDU-alignment mode on the TCP connection - else if the sender does not support PDU-alignment mode, then the receiver MAY request Marker mode from the sender. If the sender also supports Marker mode, then the sender MUST enable use of Marker mode - otherwise, the receiver has not requested use of a framing mode and the sender MUST NOT enable use of any framing mode on the TCP connection - [Note: the above rules may be moved to the Framing I-D] Interoperation: * the framing mode or modes that a receiver implementation chooses to support will determine which senders it can perform direct data placement for (since senders can choose to implemented either of the framing modes). It is anticipated that there will be receiver implementations which support the following combinations of framing modes: 1. PDU-alignment + Markers 2. PDU-alignment only 3. no framing Chunking: * Use of PDU-alignment mode requires that a dynamic chunking layer be implemented above the framing TCP layer. 3) > Still need to determine iSCSI mechanism for turning on Framing protocol Marker mode operation 4) > Still need to determine iSCSI mechanism for negotiating values for framing operational parameters 5) Perhaps there is some description of probable framing scenarios capturing the most likely combinations of the following attributes: - initiator or target - software implementation or hardware implementation - unmodified or modified TCP stack - sender AND receiver framing behaviors (no framing, or Marker mode, or PDU-Alignment mode) - values for framing operational parameters ------------------------------------------------------- The reasoning for these proposed changes is as follows: 1) iSCSI use of direct data placement and framing a) Direct data placement allows a HW-accelerated interface card to place each incoming ULP PDU (and ideally each TCP segment payload) directly into its final application buffer in end-system memory, even when the underlying TCP segments arrive out of order. The ULP PDU carries the buffer location information for doing the direct placement. This means that the TCP doesn't need to hold application data in an internal receive queue while sequence gaps are filled, nor subsequently copy that data into final application buffers. This allows the interface card to minimize or eliminate the very fast and large receive queue memory that would normally be required when running over networks with large bandwidth-delay products (10-100 Gbps, 200msec RTT). b) In order to perform direct data placement, the placement function must be able to locate ULP headers in the TCP segment stream, and extract placement information, even when TCP segments arrive out of order. A framing mechanism provides the underlying wire protocol and behaviors to enable this. c) The I-D "The Case for RDMA" (draft-csapuntz-caserdma-00.txt) discusses the benefits of direct data placement in the context of a generalized RDMA facility. 2) Merging Marker mode into "ULP Framing for TCP" I-D a) The TCP-related framing work already has mindshare in TSVWG and this work is embodied in the current framing I-D. Rather than dilute the framing effort with additional I-Ds, all framing related work should be collected into a modified version of the existing framing I-D. b) Other ULPs may also find Marker mode useful in software-only unmodified-TCP client scenarios c) The framing I-D appears to be a reasonable literary vehicle for documenting the collection of framing schemes. The I-D could be extended in the future to include a byte or word stuffing frame marker method such as COBS. d) A single framing I-D may help to encourage a single consistent interface with the ULP regardless of which framing mode is employed. e) The iSCSI spec can simply reference the one framing I-D. 3) Specifying iSCSI sender and receiver framing support a) Receivers can choose to not implement framing. Software implementations of receivers may incur extra data movements in processing framing and generally get no benefit from using framing. It is anticipated that these receiver software implementations will not support framing. b) Hardware-accelerated receivers that want to perform direct data placement and eliminate or minimize the amount of TCP reassembly memory (for links with large bandwidth-delay products) will require senders to support framing behavior. These receiver implementations are only viable if they can rely on the fact that all senders are capable of supporting some framing mode. c) Framing is done for the receiver's benefit, and is mostly a minor inconvenience for the sender. However, senders may have limitations regarding which framing mode(s) they can support. So, a sender is allowed to implement the framing mode(s) most suited to it, and the receiver is allowed to select from these supported framing modes, or choose not to utilize framing on the connection. d) There isn't one framing mode that is best for all senders: - Marker mode is best suited to software implementations that run over unmodified TCP stacks - PDU-alignment mode is best suited to hardware implementations that want to minimize or eliminate buffer memory and reduce per-packet processing complexity e) Marker mode is the best choice for a framing mechanism that can operate completely above a client TCP stack and not require any modification to that stack. - Other interval-based approaches (periodic PDU alignment, fixed length PDU) require padding and waste bandwidth - bit-stuffing and byte-stuffing schemes (COBS, etc) have a much higher processing overhead f) Marker mode supports one potentially compelling application for iSCSI involving software-only implementations on mainstream desktops and laptops operating over unmodified TCP stacks to access centralized storage arrays. These software implementations are likely to exist far into the future. Individual software-only clients may not operate at 10Gbps, but may be combined together with other clients that could aggregate to 10Gbps, thus making direct data placement compelling for a 10Gbps receiver even if the senders are only operating at 1Gbps. g) Marker mode doesn't completely eliminate the need for buffer memory on the receiver. The receiver still needs to use "eddy buffers" that temporarily hold incoming data after a dropped segment containing a ULP header up until the next ULP header is located in the packet stream, and which exist for as long as the original ULP header segment is outstanding. But Marker mode does greatly reduce the amount of memory needed as compared to a traditional TCP receiver's reassembly memory requirements (often equal to number-of-connections X round-trip-pipe-size). The Marker mode small memory requirements are dependent upon the period of the marker, and the size of the ULP PDUs being restricted to a reasonably small value. The larger that either one is, the larger the eddy buffer memory requirements. Also, an eddy buffer is required each time a ULP header is dropped, so that multiple ULP header drops in close proximity may cause multiple eddy buffers to be temporarily pending on a connection. h) PDU-alignment framing mode allows each ULP PDU to be aligned with, and sent in, a single TCP segment under normal conditions (with the added requirement that a chunking layer needs to be implemented between iSCSI and the framing TCP stack). This behavior allows each TCP segment to be fully self-describing with respect to direct placment. Thus, each incoming TCP segment payload can be processed and direct placed as it arrives with no residual state information nor eddy buffer memory required. i) PDU-alignment framing mode does require the use of a small number of "eddy buffers" when dynamic changes in the network path MTU occur and packets arrive out of order. j) The PDU-alignment framing mode is preferred. However, it may be several years before all of the different software TCP/IP implementations will be able to support framing behavior. To do this, software interfaces will need to change, and something needs to drag them there. This will take some time, so we have markers to help out in the meantime. If all receivers that can use framing can do either one, and senders that can do PDU-alignment should do so, we will have a larger set of PDU-alignment implementations that may help pull the rest of the software interfaces along with the k) It is anticipated that a receiver will not implement only Markers. The receiver implementations will probably be: 1. PDU-alignment + Markers 2. PDU-alignment only 3. no framing ----------------------------------------------------------- Open Issues: 1) Interoperability between sender and receiver - Given that both senders and receivers have a choice in which framing mode(s) they implement, there is the potential for the sender to implement one framing mode and the receiver to implement a different framing mode (e.g. the sender implements only Marker mode, and the receiver implements only PDU-alignment mode). - In this situation, the receiver and sender would not enable framing on the TCP connection, and the receiver would not be able to perform direct data placement. Throughput from sender to receiver would likely be greatly reduced should any TCP segment drops occur. 2) None of the current framing schemes take TCP data integrity into account. It either needs to be decided: a) how to detect when a data integrity problem occurs within a framing header, and what to do about it (even if it just kills the TCP connection), b) or that a sufficient level of data integrity needs to be provided for all protocols running over TCP via a more holistic approach. 3) Acceptability of the PDU-Alignment framing mode's reliance on "key+length" matching across resegmenting middleboxes - In PDU-Alignment mode each TCP segment payload contains one complete framing PDU (consisting of an 8 byte framing header followed by one or more complete ULP PDUs). Thus, every TCP segment has the TCP header followed immediately by the framing header. - In certain cases a single framing PDU must be broken across multiple TCP segments (such as dynamic Path MTU reductions), resulting in TCP segments where a framing header doesn't immediately follow the TCP header. - The framing I-D defines sender behaviors that allow PDU-alignment mode to function deterministically and correctly in all cases where the TCP segmentation flowing from sender to receiver is not altered. - If the TCP segmentation from sender to receiver is altered by an intermediary (resegmenting middlebox), and a framing-header-containing segment drop or reordering has occurred such that the receiver is attempting to locate the next framing header in the segment stream, then the receiver must examine the first 8 bytes of each incoming TCP segment payload for a valid framing header containing valid Key(6B) and Length(2B) fields. - A false-positive occurs if, upon resegmentation by a middlebox, the receiver gets a TCP segment in which the first 8 bytes of the payload indicate a valid framing header (the first 6 bytes match the previously exchanged random key value, and the next 2 bytes contain a valid length), yet the TCP segment payload isn't actually a framing header. - While it is felt that the probability of a false-positive in these resegmenting-middlebox scenarios will be sufficiently low, further analysis work may be may be required in this area. - Note that this mechanism is NOT a scanning technique for locating start-of-frame across an arbitrary byte stream. It only provides an indication of PDU alignment or not. The first 8 bytes of the TCP segment payload are examined to determine if the segment contains the start of a ULP PDU. 4) Do Markers work at 10Gbps? - The feasibility of markers at 10Gbps has been questioned. It would be beneficial to hear specifics regarding why Markers won't work at 10Gbps. Markers don't allow for a no-memory direct data placement NIC since eddy-buffers are required. So, support for clients with unmodified TCP stacks comes at a cost, which is the cost of supporting eddy buffers on the NIC. - One question is whether the eddy buffers can be contained entirely in the ASIC or need to be in off-chip memory. ----------------------------------------------------------
Home Last updated: Tue Sep 04 01:04:08 2001 6315 messages in chronological order |