[openib-general] QoS RFC - Resend using a friendly mailer
Michael Krause
krause at cup.hp.com
Tue May 30 11:01:36 PDT 2006
High-level feedback:
- An IB fabric could be used for a single ULP and still require QoS. The
issue is how to differentiate flows on a given shared element within the
fabric.
- QoS controls must be dynamic. The document references initialization as
the time when decisions are made but obviously that is just a first pass on
use of the fabric and not what it will become in potentially a short period
of time.
- QoS also involves multi-path support (not really touched upon in terms of
specifics in this document). Distributing or segregating work even if for
the same ULP should be done across multiple or distinct paths. In one
sense this may complicate the work but in another it is simpler in that
arbitration controls for shared links become easier to manage if the number
of flows is reduced.
- IP over IB defines a multicast group which is ultimately a spanning
tree. That should not constrain what paths are used to communicate between
endnode pairs. That only defines the multicast paths which are not
strongly ordered relative to the unicast traffic. Further IP over IB may
operate using the RC mode between endnodes. It is very simple to replicate
RC and then segregate these into QoS domains (one could just align priority
with the 802.1p for simplicity and practical execution) which can in turn
flow over shared or distinct paths.
- IB is a centrally managed fabric. Adding in SID into records and such
really isn't going to help solve the problem unless there is also a
centralized management entity well above IB that can prioritize
communication service rates for different ULP and endnode pairs. Given
most of these centralized management entities are rather ignorant of IB at
the moment, this presents a chicken-egg dilemma which is further
complicated by developing SOA technology. It might be more valuable in one
sense to examine SOA technology and how it is translating itself to say
Ethernet and then see how this can be leveraged to IB.
- QoS needs to examine the sums of the consumers of a given path and their
service rate requirements. It isn't just about setting a priority level
but also about the packet injection rate to the fabric on that
priority. This needs to be taken into account as well.
Overall, it is not clear to me what the end value of this document. The
challenge for any network admin is to translate SOA driven requirements
into fabric control knob setting. Without such translation algorithms /
understanding, it is not clear that there is anything truly missing in the
IBTA spec suite or that this RFC will really advance the integration of IB
into the data center in a truly meaningful manner.
Mike
At 07:53 AM 5/30/2006, Eitan Zahavi wrote:
>To: OPENIB <openib-general at openib.org>
>Subject: QoS RFC - Resend using a friendly mailer
>--text follows this line--
>Hi All
>
>Please find the attached RFC describing how QoS policy support could be
>implemented in the OpenFabrics stack.
>Your comments are welcome.
>
>Eitan
>
> RFC: OpenFabrics Enhancements for QoS Support
> ===============================================
>
>Authors: . Eitan Zahavi <eitan at mellanox.co.il>
>Date: .... May 2006.
>Revision: 0.1
>
>Table of contents:
>1. Overview
>2. Architecture
>3. Supported Policy
>4. CMA functionality
>5. IPoIB functionality
>6. SDP functionality
>7. SRP functionality
>8. iSER functionality
>9. OpenSM functionality
>
>1. Overview
>------------
>Quality of Service requirements stem from the realization of I/O
>consolidation
>over IB network: As multiple applications and ULPs share the same fabric,
>means
>to control their use of the network resources are becoming a must. The basic
>need is to differentiate the service levels provided to different traffic
>flows.
>Such that a policy could be enforced and control each flow utilization of the
>fabric resources.
>
>IBTA specification defined several hardware features and management
>interfaces
>to support QoS:
>* Up to 15 Virtual Lanes (VL) could carry traffic in a non-blocking manner
>* Arbitration between traffic of different VL is performed by a 2 priority
> levels weighted round robin arbiter. The arbiter is programmable with
> a sequence of (VL, weight) pairs and maximal number of high priority
> credits
> to be processed before low priority is served
>* Packets carry class of service marking in the range 0 to 15 in their
> header SL field
>* Each switch can map the incoming packet by its SL to a particular output
> VL based on programmable table VL=SL-to-VL-MAP(in-port, out-port, SL)
>* The Subnet Administrator controls each communication flow parameters
> by providing them as a response to Path Record query
>
>The IB QoS features provide the means to implement a DiffServ like
>architecture.
>DiffServ architecture (IETF RFC2474 2475) is widely used today in highly
>dynamic
>fabrics.
>
>This proposal provides the detailed functional definition for the various
>software elements that are required to enable a DiffServ like architecture
>over
>the OpenFabrics software stack.
>
>
>
>
>
>2. Architecture
>----------------
>This proposal split the QoS functionality between the SM/SA, CMA and the
>various
>ULPS. We take the "chronology approach" to describe how the overall system
>works:
>
>2.1. The network manager (human) provides a set of rules (policy) that
>defines
>how the network is being configured and how its resources are split to
>different
>QoS-Levels. The policy also define how to decide which QoS-Level each
>application or ULP or service use.
>
>2.2. The SM analyzes the provided policy to see if it is realizable and
>performs
>the necessary fabric setup. The SM may continuously monitor the policy and
>adapt
>to changes in it. Part of this policy defines the default QoS-Level of each
>partition. The SA is being enhanced to match the requested Source,
>Destination,
>TClass, Service-ID (and optionally SL and priority) against the policy. So
>clients (ULPs, programs) can obtain a policy enforced QoS. The SM is also
>enhanced to support setting up partitions with appropriate IPoIB broadcast
>group. This broadcast group carries its QoS attributes: TClass, SL, MTU and
>RATE.
>
>2.3. IPoIB is being setup. IPoIB uses the SL, MTU and RATE available on the
>multicast group which forms the broadcast group of this partition.
>
>2.4. MPI which provides non IB based connection management should be
>configured
>to run using hard coded SLs. It uses these SLs in every QP being opened.
>
>2.5. ULPs that use CM interface (like SRP) should have their own pre-assigned
>Service-ID and use it while obtaining PathRecord for establishing their
>connections. The SA receiving the PathRecord should match it against the
>policy
>and return the appropriate PathRecord including SL, MTU, RATE and TClass.
>
>2.6. ULPs and programs using CMA to establish RC connection should provide
>the
>CMA the target IP and Service-ID. Some of the ULPs might also provide TClass
>(E.g. for SDP sockets that are provided the TOS socket option). The CMA
>should
>then use the provided Service-ID and optional TClass and pass them in the
>PathRecord request. The resulting PathRecord should be used for
>configuring the
>connection QP.
>
>PathRecord and MultiPathRecord enhancement for QoS:
>As mentioned above the PathRecord and MultiPathRecord attributes should be
>enhanced to carry the Service-ID which is a 64bit value. Given the existing
>definition for these attributes we propose to use the following fields for
>Service-ID:
>* For PathRecord: use the first 2 reserved fields whicg are 32bits each
> (component masks 0x1 and 0x2). Component mask 1 should be used to refer
> to the
> merged Service-ID field
>* For MultiPathRecord: use 2 reserved fields:
> 1. after the packet life (8 bits) which is component mask bit 0x10000 (17)
> 2. the field before SDGID1 (56 bits) which is component mask bit
> 0x200000 (22)
> Once merged they should be selected using component mask bit 0x10000 (17)
>A new capability bit should describe the SM QoS support in the SA class port
>info. This approach provides an easy migration path for existing access layer
>and ULPs by not introducing a new attribute.
>
>
>3. Supported Policy
>--------------------
>
>The QoS policy supported by this proposal is divided into 4 sub sections:
>
>* Node Group: a set of HCAs, Routers or Switches that share the same
>settings.
>A node groups might be a partition defined by the partition manager policy in
>terms of GUIDs. Future implementations might provide support for
>NodeDescription
>based definition of node groups.
>
>* Fabric Setup:
>Defines how the SL2VL and VLArb tables should be setup. This policy
>definition
>assumes the computation of target behavior should be performed outside of
>OpenSM.
>
>* QoS-Levels Definition:
>This section defines the possible sets of parameters for QoS that a client
>might
>be mapped to. Each set holds: SL and optionally: Max MTU, Max Rate, Path Bits
>(in case LMC > 0 is used for QoS) and TClass.
>
>* Matching Rules:
>A list of rules that match an incoming PathRecord request to a QoS-Level. The
>rules are processed in order such as the first match is applied. Each rule is
>built out of set of match expressions which should all match for the rule to
>apply. The matching expressions are defined for the following fields
>** SRC and DST to lists of node groups
>** Service-ID to a list of Service-ID or Service-ID ranges
>** TClass to a list of TClass values or ranges
>
>XML style syntax is provided for the policy file. However, a strict BNF
>format
>(provided in section 8) should be used for parsing it.
>
><?xml version="1.0" encoding="ISO-8859-1"?>
><qos-policy>
> <!-- Port Groups define sets of ports to be used later in the settings -->
> <port-groups>
> <!-- using port GUIDs -->
> <port-group> <name>Storage</name> <use>our SRP storage targets</use>
> <port-guid>0x1000000000000001</port-guid>
> <port-guid>0x1000000000000002</port-guid>
> </port-group>
> <!-- using names obtained by concatenation of first 2 words of
> NodeDescription
> and port number -->
> <port-group> <name>Virtual Servers</name> <use>node desc and IB port
> #</use>
> <port-name>vs1/HCA-1/P1</port-name>
> <port-name>vs3/HCA-1/P1</port-name>
> <port-name>vs3/HCA-2/P1</port-name>
> </port-group>
> <!-- using partitions defined in the partition policy -->
> <port-group> <name>Partition 1</name> <use>default settings</use>
> <partition>Part1</partition>
> </port-group>
> <!-- using node types HCA|ROUTER|SWITCH -->
> <port-group> <name>Routers</name> <use>all routers</use>
> <node-type>ROUTER</node-type>
> </port-group>
> </port-groups>
> <qos-setup>
> <!-- define all types of SL2VL tables always have 16 VL entries -->
> <sl2vl-tables>
> <!-- scope defines the exact devices and in/out ports the tables apply to
> if the same port is matching several rules the last one applies -->
> <sl2vl-scope> <group>Part1</group> <from>*</from> <to>*</to>
> <sl2vl-table>0,1,2,3,4,5,6,7,8,9,10,11,12,13,14</sl2vl-table>
> </sl2vl-scope>
> <!-- "across" means the port just connected to the given group,
> also the link across port 1 is probably supporting only 2 VLs -->
> <sl2vl-scope> <across>Storage</across> <from>*</from> <to>1</to>
> <sl2vl-table>0,1,1,1,1,1,1,1,1,1,1,1,1,1,1</sl2vl-table>
> </sl2vl-scope>
> <sl2vl-tables>
>
> <!-- define all types of VLArb tables. The length of the tables should
> match the physically supported tables by their target ports -->
> <vlarb-tables>
> <!-- scope defines the exact ports the VLArb tables apply to -->
> <vlarb-scope> <group>Storage</group> <to>*</to>
> <!-- VLArb table holds VL and weight pairs -->
> <vlarb-high>0:255,1:127,2:63,3:31,4:15,5:7,6:3,7:1</vlarb-high>
> <vlarb-low>8:255,9:127,10:63,11:31,12:15,13:7,14:3</vlarb-low>
> <vl-high-limit>10</vl-high-limit>
> </vlarb-scope>
> </vlarb-tables>
> </qos-setup>
>
><qos-levels>
> <!-- the first one is just setting SL -->
> <qos-level> <sn>1</sn> <use>for the lowest priority comm</use>
> <sl>16</sl>
> </qos-level>
> <!-- the second sets SL and TClass -->
> <qos-level> <sn>2</sn> <use>low latency best bandwidth</use>
> <sl>0</sl> <tclass>7</tclass>
> </qos-level>
> <!-- the whole set: SL, TClass, MTU-Limit, Rate-Limit, Path-Bits -->
> <qos-level> <sn>3</sn> <use>just an example</use>
> <sl>0</sl> <tclass>32</tclass> <mtu_limit>1</mtl_limit>
> <rate_limit>1</rate_limit>
> </qos-level>
> </qos-levels>
>
> <qos_match_rules>
> <!-- matching by single criteria: tclass (list of values and ranges) -->
> <qos_match_rule> <sn>1</sn> <use>low latency by tclass 7-9 or 11></use>
> <tclass>7-9,11</tclass> <match-level>1</match-level>
> </qos_match_rule>
> <!-- show matching by destination group AND service-ids -->
> <qos_match_rule> <sn>2</sn> <use>Storage targets connection></use>
> <destination>Storage</destination> <service>22,4719</service>
> <match-level>3</match-level>
> </qos_match_rule>
> </qos_match_rules>
>
></qos-policy>
>
>
>4. IPoIB
>---------
>
>IPoIB already query the SA for its broadcast group information. The
>additional
>functionality required is for IPoIB to provide the broadcast group SL,
>MTU, RATE
>and TClass in every following PathRecord query performed when a new UDAV is
>needed by IPoIB.
>We could assign a special Service-ID for IPoIB use but since all
>communication
>on the same IPoIB interface shares the same QoS-Level without the ability to
>differentiate it by target service we can ignore it for simplicity.
>
>5. CMA features
>----------------
>
>The CMA interface supports Service-ID through the notion of port space as a
>prefixes to the port_num which is part of the sockaddr provided to
>rdma_resolve_add(). What is missing is the explicit request for a TClass that
>should allow the ULP (like SDP) to propagate a specific request for a
>class of
>service. A mechanism for providing the TClass is available in the IPv6
>address,
>so we could use that address field. Another option is to implement a special
>connection options API for CMA.
>
>Missing functionality by CMA is the usage of the provided TClass and
>Service-ID
>in the sent PathRecord. When a response is obtained it is an existing
>requirement for the CMA to use the PathRecord from the response in setting up
>the QP address vector.
>
>
>6. SDP
>-------
>
>SDP uses CMA for building its connections.
>The Service-ID for SDP is 0x000000000001PPPP, where PPPP are 4 hex digits
>holding the remote TCP/IP Port Number to connect to.
>SDP might be provided with SO_PRIORITY socket option. In that case the value
>provided should be sent to the CMA as the TClass option of that connection.
>
>7. SRP
>-------
>
>Current SRP implementation uses its own CM callbacks (not CMA). So SRP should
>fill in the Service-ID in the PathRecord by itself and use that
>information in
>setting up the QP. The T10 SRP standard defines the SRP Service-ID to be
>defined
>by the SRP target I/O Controller (but they should also comply with IBTA
>Service-
>ID rules). Anyway, the Service-ID is reported by the I/O Controller in the
>ServiceEntries DMA attribute and should be used in the PathRecord if the SA
>reports its ability to handle QoS PathRecords.
>
>8. iSER
>--------
>iSER uses CMA and thus should be very close to SDP. The Service-ID for iSER
>should be TBD.
>
>
>9. OpenSM features
>-------------------
>The QoS related functionality to be provided by OpenSM can be split into two
>main parts:
>
>3.1. Fabric Setup
>During fabric initialization the SM should parse the policy and apply its
>settings to the discovered fabric elements. The following actions should be
>performed:
>* Parsing of policy
>* Node Group identification. Warning should be provided for each node not
> specified but found.
>* SL2VL settings validation should be checked:
> + A warning will be provided if there are no matching targets for the
> SL2VL
> setting statement.
> + An error message will be printed to the log file if an invalid
> setting is
> found. A setting is invalid if it refers to:
> - Non existing port numbers of the target devices
> - Unsupported VLs for the target device. In the later case the map to non
> existing VLs should be replaced to VL15 i.e. packets will be dropped.
>* SL2VL setting is to be performed
>* VL Arbitration table settings should be validated according to the
>following
> rules:
> + A warning will be provided if there are no matching targets for the
> setting
> statement
> + An error will be provided if the port number exceeds the target ports
> + An error will be generated if the table length exceeds device
> capabilities
> + An warning will be generated if the table quote a VL that is not
> supported
> by the target device
>* VL Arbitration tables will be set on the appropriate targets
>
>3.2. PathRecord query handling:
>OpenSM should be able to enforce the provided policy on client request.
>The overall flow for such requests is: first the request is matched
>against the
>defined match rules such that the target QoS-Level definition is found. Given
>the QoS-Level a path(s) search is performed with the given restrictions
>imposed
>by that level. The following two sections describe these steps.
>
>One issue not standardized by the IBTA is how Service-ID is carried in the
>PathRecord and MultiPathRecord attributes. There are basically two options:
>a. Replace the SM-Key field by the Service-ID. In that case no
>component mask
> bit will be assigned to it. Such that if the field is zero we should
> treat it
> as if the component mask bit is clear.
>b. Encode it into spare fields. For PathRecord the first two fields are
>reserved
> and are 64 bit when combined. The first component mask bit maps to the
> first
> reserved field and should be used for Service-ID masking. For
> MultiPathRecord
> attribute there are no adjacent reserve fields that makes a 64 bit
> field. So
> the reserve field following the packet-lifetime (8 bits) combined with
> the
> reserved field DGIDCount (56 bits) can make the Service-ID. In this
> case also
> the first reserve field component mask bit should be used as the
> Service-ID
> component mask bit.
>
>
>
>3.2.1. Matching rule search:
>A rule is "matching" a PathRecord request using the following criteria:
>* Matching rules provide values in a list of either single value, or range
>of
> values. A PathRecord field is "matching" the rule field if it is
> explicitly
> noted in the list of values or is one of the values covered by a range
> included in the field values list.
>* Only PathRecord fields that have their component mask bit set should be
> compared.
>* For a rule to be "matching" a PathRecord request all the rule fields
>should be
> "matching" their PathRecord fields. Such that a PathRecord request that
> does
> not have a component mask field set for one of the rule defined
> fields can
> not match that rule.
>* A PathRecord request that have a component mask bit set for one of the
>fields
> that is not defined by the rule can match the rule.
>
>The algorithm to be used for searching for a rule match might be as simple
>as a
>sequential search through all rules or enhanced for better performance. The
>semantics of every rule field and its matching PathRecord field are described
>below:
>* Source: the SGID or SLID should be part of this group
>* Destination: the DGID or DLID should be part of this group
>* Service-ID: check if the requested Service-ID (available in the
>PathRecord old
> SM-Key field) is matching any of this rule Service-IDs
>* TClass: check if the PathRecord TClass field is matching
>
>3.2.2 PathRecord response generation:
>The QoS-Level pointed by the first rule that matches the PathRecord request
>should be used for obtaining the response SL, MTU-Limit, RATE-Limit,
>Path-Bits
>and TClass. A default QoS-Level should be used if no rule is matching the
>query.
>
>The efficient algorithm for finding paths that meet the QoS-Level criteria is
>beyond the scope of this RFC and left for the implementer to provide. However
>the criteria by which the paths match the QoS-Level are described below:
>
>* SL: The paths found should all use the given SL. For that sake PathRecord
> algorithm should traverse the path from source to destination only through
> ports that carry a valid VL (not VL15) by the SL2VL map (should
> consider input
> and output ports and SL).
>* MTU-Limit: The resulting paths MTU should not exceed the given MTU-Limit
>* Rate-Limit: The resulting paths RATE should not exceed the given RATE-Limit
> (rate limit is given in units of link BW = Width*Speed according to IBTA
> Specification Vol-1 table-205 p-901 l-24).
>* Path-Bits: define the target LID lowest bits (number of bits defined by the
> target port PortInfo.LMC field). The path should traverse the LFT using
> the
> target port LID with the path-bits set.
>* TClass: should be returned in the result PathRecord. When routing is
>going to
> be supported by OpenSM we might use this field in selecting the target
> router too in a TBD way.
>
>_______________________________________________
>openib-general mailing list
>openib-general at openib.org
>http://openib.org/mailman/listinfo/openib-general
>
>To unsubscribe, please visit
>http://openib.org/mailman/listinfo/openib-general
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20060530/151b4c88/attachment.html>
More information about the general
mailing list