Switch draft Q&A
Kevin Dubray <kdubray@baynetworks.com> Mon, 28 July 1997 22:48 UTC
Received: from cnri by ietf.org id aa23842; 28 Jul 97 18:48 EDT
Received: from mailhost3.BayNetworks.COM (ns4.BayNetworks.COM [192.32.253.7]) by cnri.reston.va.us (8.8.5/8.7.3) with ESMTPid SAA19140 for <ietfadm@cnri.reston.va.us>; Mon, 28 Jul 1997 18:47:07 -0400 (EDT)
Received: from pobox.engeast.BayNetworks.COM (pobox.corpeast.baynetworks.com [192.32.151.199]) by mailhost3.BayNetworks.COM (8.8.6/BNET-97/07/07-E) with ESMTP id SAA18890
Received: (majordom@localhost) by pobox.engeast.BayNetworks.COM (SMI-8.6/BNET-97/04/24-S) id SAA06152; Mon, 28 Jul 1997 18:00:59 -0400 for bmwg-outgoing
Received: from mailhost1.BayNetworks.COM (southpass.corpwest.baynetworks.com [134.177.3.16]) by pobox.engeast.BayNetworks.COM (SMI-8.6/BNET-97/04/24-S) with ESMTP id SAA06146; Mon, 28 Jul 1997 18:00:58 -0400 for <bmwg@pobox.engeast>
Received: from pobox.engeast.BayNetworks.COM (pobox.corpeast.baynetworks.com [192.32.151.199]) by mailhost1.BayNetworks.COM (8.8.6/BNET-97/07/07-E) with ESMTP id PAA10331; Mon, 28 Jul 1997 15:00:55 -0700 (PDT) for <bmwg@baynetworks.com>
Posted-Date: Mon, 28 Jul 1997 15:00:55 -0700 (PDT)
Received: from pestilence.engeast (pestilence [192.32.180.194]) by pobox.engeast.BayNetworks.COM (SMI-8.6/BNET-97/04/24-S) with SMTP id SAA06142; Mon, 28 Jul 1997 18:00:55 -0400 for <bmwg@baynetworks.com>
Date: Mon, 28 Jul 1997 18:00:55 -0400
From: Kevin Dubray <kdubray@baynetworks.com>
Message-Id: <199707282200.SAA06142@pobox.engeast.BayNetworks.COM>
To: bmwg@baynetworks.com
Subject: Switch draft Q&A
Sender: owner-bmwg@baynetworks.com
Precedence: bulk
Hi Scott and all, I have given your remarks a lot of consideration as I think you will see from the amendments I have made to the document. I have commented on your comments and annotated them in the reply message below. Each of my comments starts with '<Bob replies . . .' and ends with '>'. In general I think that you will find that your comments have led me to say things more simply and, I hope, more clearly. In a rush of optimism I have taken out the time to 'freeze' the document by formatting it as per the Guideline RFC. Thanks for the comments and have a good read. Bob [from Scott] Hi Bob & all I'm (belatedly) taking a look at draft-ietf-bmwg-lanswitch-05.txt and have some comments --------------- 3.1.1 Device under test (DUT) Discussion: A single stand-alone or modular unit generally equipped with its own power supply. -- I don't think the reference to a power supply is needed I'd say something like: A single stand-alone or modular unit which forwards frames of data from one or more interfaces to one or more interfaces. < Bob replies: Agreed. I have reworked the discussion and included what I think is an essential reference to addressing information. Discussion: A single stand-alone or modular unit which receives frames on or more of its interfaces and then forwards them to one or more interfaces according to the addressing information contained in the frame. > -------------- 3.2.1 Unidirectional traffic Definition: Frames presented to a DUT/SUT such that the receiving and transmitting interfaces are mutually exclusive. 3.3.2 Partially meshed traffic Definition: Frames forwarded between mutually exclusive sets of input and output interfaces of a DUT/SUT. -- I'm a bit confused by this pair of definitions - they do not seem to say something different - I'm not quite sure how to change somthing to fix the definitions - in general the 3.2 section and the 3.3 sections seem to overlap and need to have some reworking and the discussions do not clarify things for me - I suggest that the Partially meshed traffic def & discussion should be worked on to clarify things <Bob replies: to deal with your concerns (which I share) I have first cast the definition of unidirectional traffic in different terms to underline that this defintion has to do with traffic orientation. It now reads: 3.2.1 Unidirectional traffic Definition:When all frames presented to the input interfaces of a DUT/SUT are addressed to output interfaces which do not themselves receive any frames. I have also put the following paragraph to clarify the discussion section: When a tester offers unidirectional traffic to a DUT/SUT reception and transmission are handled by different interfaces or sets of interfaces of the DUT/SUT. All frames received from the tester by the DUT/SUT are transmitted back to the tester from interfaces which do not themselves receive any frames. As for Partially meshed traffic (3.3.2) I have modified the definition to better reflect that we are talking about traffic distribution: 3.3.2 Partially meshed traffic Definition: Frames offered to one or more input interfaces of a DUT/SUT and addressed to one or more output interfaces where input and output interfaces are mutually exclusive and mapped one-to-many, many-to-one or many-to-many. I have also modified the discussion on partially meshed traffic to show how it ties in with the terms unidirectional and bidirectional traffic. It now reads: When partially meshed traffic is distributed in a one-to-many or many-to-one mapping of receiving to transmitting interfaces of a DUT/SUT traffic orientation will be unidirectional. When traffic is partially meshed and distributed in a many-to-many mapping of receiving to transmitting ports which are mutually exclusive traffic orientation will be bidirectional. > ----------------- 3.2.1 Unidirectional traffic Discussion: I think the discussion could be clarified to some degree - I took a cut at part of it It is useful to distinguish traffic orientation and traffic distribution when considering traffic patterns used in device testing. Unidirectional traffic, for example, is traffic orientated in a single direction between mutually exclusive sets of source and destination interfaces of a DUT/SUT. Such traffic, however, can be distributed between interfaces in different ways. When traffic is sent to two or more interfaces from an external source and then forwarded by the DUT/SUT to a single output interface the traffic orientation is unidirectional and the traffic distribution between interfaces is many-to-one. Traffic can also be sent to a single input interface and forwarded by the DUT/SUT to two or more output interfaces to achieve a one-to-many distribution of traffic. <Bob replies: Agreed: wording got to be somewhat abstruce as attempts were made to "tune" this discussion. Your paragraph reads much better so I have pasted it in as is. In addition I have added the new paragraph just above: "When a tester . . .". > -- Such traffic distributions can also be combined to test for head of line blocking or to measure forwarding rates and throughput when congestion control is active. -- the See Also section should include 3.7.3 Head of line blocking and 3.7 Congestion control <Bob replies: agreed and done See Also: bidirectional traffic (3.2.2) one-to-one mapped traffic (3.3.1) partially meshed traffic (3.3.2) fully meshed traffic (3.3.3) congestion control (3.7) head of line blocking (3.7.3) > ------------------ 3.2.2 Bidirectional traffic Definition: Frames presented to a DUT/SUT such that the interfaces of the DUT/SUT both receive and transmit. -- I'd change just a bit Frames presented to a DUT/SUT such that each of the interfaces of the DUT/SUT both receive and transmit. <Bob replies: agreed and done with your suggested wording: Frames presented to a DUT/SUT such that each of the interfaces of the DUT/SUT both receive and transmit. I have also the following sentence to the discussion section to parallel the addition to the unidirectional discussion: When a tester offers bidirectional traffic to a DUT/SUT all the interfaces which receive frames from the tester also transmit frames back to the tester. > ------------------- 3.2.2 Bidirectional traffic Discussion: This definition conforms to the discussions in sections 14 and 16 of RFC 1944 on bidirectional traffic and multi-port testing. Bidirectional traffic MUST be offered when measuring throughput on full duplex interfaces of a switching device. -- all RFCs wich use the terms like "MUST" "MAY" etc should now contain a reference to RFC 2119 - the exact form of the reference is in the RFC <Bob replies: thanks for pointing this out. I have put the reference in as per RFC 2119 in the Existing Definitions section (2) right after the Introduction (1) and just before the Term Definitions (3). > -------------------- 3.3.3 Fully meshed traffic It should be noted that bidirectional multi-port traffic can load backbone connections linking together two switching devices more than meshed traffic. -- it would be nice to expand on this a bit <Bob replies - I have already had two "off-line" requests making the same point so I have expanded the discussion as follows: [by the by: I have received a large number of comments on the draft at various stages but their authors seem consistently to shy away from posting them on the list despite Kevin Dubray's repeated and vigorous call for people to communicate their comments via the appropriate channel. Unfortunately this habit does not allow the list to reflect the considerable amount of commenting that goes on as a draft such as this one gets elaborated.] It should be noted that bidirectional multi-port traffic can load backbone connections linking together two switching devices more than fully meshed traffic. In a bidirectional multiport test each one of two devices or systems connected over a backbone connection can be configured to forward the totality of the frames they receive to the devices or systems placed on the opposite side of the backbone connection. All frames generated during such a test will traverse the backbone connection. Fully meshed traffic requires at least some of the frames received on the interfaces of each one of the devices or systems under test to be forwarded locally, that is to the interfaces of the receiving devices themselves. Such frames will not traverse the backbone connection. > ---------------------- 3.3.3 Fully meshed traffic Bidirectional meshed traffic on half duplex interfaces is inherently bursty since interfaces must interrupt transmission whenever they receive frames. This kind of bursty meshed traffic is characteristic of real network traffic and can be advantageously used to diagnose a DUT/SUT by exercising many of its component parts simultaneously. Additional inspection may be warranted to correlate the frame forwarding capacity of a DUT/SUT when offered meshed traffic and the behavior of individual elements such as input or output buffers, buffer allocation mechanisms, aggregate switching capacity, processing speed or medium access control. -- it might be good to add that analysis of the results of such tests can present a chalenge since you need to take into account the actual rate at which data was presented to the DUT/SUT rather than just the rate that the tester attempted to send. (i.e. the offered vs intended load) - the see also should point to them <Bob replies: Agreed. Here is what I have added: The analysis of forwarding rate measurements presents a challenge when offering bidirectional or fully meshed traffic since the rate at which the tester can be observed to transmit frames to the DUT/SUT may be smaller than the rate at which it intends to transmit due to collisions on half duplex media or the action of congestion control mechanisms. This makes it important to take account of both the intended and offered loads defined in sections 3.5.1.and 3.5.2 below when reporting the results of such forwarding rate measurements. The See Also now reads: See Also: unidirectional traffic (3.2.1) bidirectional traffic (3.2.2) one-to-one mapped traffic (3.3.1) partially meshed traffic (3.3.2) burst (3.4.1) intended load (3.5.1) [ADDED] offered load (3.5.2) [ADDED] > ----------------------- 3.4.2 Burst size Discussion: Burst size can range from one to infinity. In unidirectional traffic there is no theoretical limit to burst length. -- I'd tweak this a bit Burst size can range from one to infinity. In unidirectional traffic and in bidirectional or meshed bursts on full duplex interfaces there is no theoretical limit to burst length. <Bob replies: fine point. I have made the adjustment. The text now reads: Burst size can range from one to infinity. In unidirectional traffic as well as in bidirectional or meshed traffic on full duplex interfaces there is no theoretical limit to burst length. > ----------------------- 3.4.3 Inter-burst gap (IBG) Bidirectional and meshed traffic are inherently bursty since interfaces share their time between receiving and transmitting frames. External sources offering bursty traffic for a given frame size and burst size must adjust the inter-burst gap to achieve a specified rate of transmission. -- I'd tewak this a bit External sources offering bursty traffic for a given frame size and burst size must adjust the inter-burst gap to achieve a specified average rate of frame transmission. <Bob replies: point well taken. Done. Text now reads: Bidirectional and meshed traffic are inherently bursty since interfaces share their time between receiving and transmitting frames. External sources offering bursty traffic for a given frame size and burst size must adjust the inter-burst gap to achieve a specified average rate of frame transmission. > ----------------------- 3.5.1 Intended load (Iload) In the case of Ethernet an external source of traffic must implement the truncated binary exponential back-off algorithm to ensure that it is accessing the medium legally. -- I'd change the "must" to "MUST" <Bob replies: yes, done. Text now reads: In the case of Ethernet an external source of traffic MUST implement the truncated binary exponential back-off algorithm to ensure that it is accessing the medium legally. > the See Also should include references to the burst & interburst gap sections <Bob replies: done. See Also: burst (3.4.1) [ADDED] inter-burst gap (3.4.3) [ADDED] offered load (3.5.2) > ------------------------ 3.5.2 Offered load (Oload) I'd also note that analysis of testing where offered ~= intended must that the difference into account <Bob replies: Agreed. This need be pointed out here. I have added bidirectional,fully meshed traffic and forwarding rate to the See Also section where the same point is raised. Here is the wording I've put into 3.5.2.: The load which an external device can be observed to apply to a DUT/SUT may be less than the intended load due to collisions on half duplex media or the action of congestion control mechanisms. This makes it important to distinguish intended and offered load when analyzing the results of forwarding rate measurements using bidirectional or fully meshed traffic. > ----------------------- 3.5.4 Overloading I'd add that information obtained through this type of test can present opportunities for abuse by marketeers <Bob replies: Picking up on your lead this is what I have added: An external source can also overload an interface by transmitting frames below the minimum inter-frame gap. A DUT/SUT MUST forward such frames at intervals equal to or above the minimum gap specified in standards. A DUT/SUT using congestion control techniques such as backpressure or forward pressure may exhibit no frame loss when a tester attempts to overload one or more of its interfaces. This should not be exploited to suggest that the DUT/SUT supports rates of transmission in excess of the maximum rate allowed by the medium since both techniques reduce the rate at which the tester offers frames to prevent overloading. Backpressure achieves this purpose by jamming the transmission interfaces of the tester and forward pressure by hindering the tester from gaining fair acces to the medium. Analysis of both cases should take the distinction between intended load (3.5.1) and offered load (3.5.2) into account. I have also made the appropriate additions to the See Also: See Also: See Also: unidirectional traffic (3.2.1) intended load (3.5.1) [ADDED] offered load (3.5.2) [ADDED] forwarding rate (3.6.1) [ADDED] backpressure (3.7.1) [ADDED] forward pressure (3.7.2) [ADDED] > ----------------------- 3.7 Congestion control I'd also note the dificulity of understanding performance results in devices that include congestion control <Bob replies: I have added two paragraphs, one in the backpressure, the other in the forward pressure section to point out the pitfalls encountered when trying to make sense of performance results on devices using one and the other congestion control techniques. I've added the following to the backpressure section( 3.7.1): A DUT/SUT applying backpressure may exhibit no frame loss when a tester attempts to overload one or more of its interfaces. This should not be interpreted to suggest that the interfaces of the DUT/SUT support forwarding rates above the maximum rate allowed by the medium. In such cases overloading is only apparent since through the application of backpressure the DUT/SUT attenuates the offered load by jamming one or more of the interfaces of the tester. And added tothe See Also: See Also: intended load (3.5.1) [ADDED] offered load (3.5.2) [ADDED] overloading (3.5.4) [ADDED] forwarding rate (3.6.1) [ADDED] forward pressure (3.7.2) I have also added the following to the forward pressure section (3.7.2) A DUT/SUT applying forward pressure may eliminate all or most frame loss when a tester attempts to overload one or more of its interfaces. This should not be interpreted to suggest that the interfaces of the DUT/SUT can sustain forwarding rates above the maximum rate allowed by the medium. Overloading in such cases is only apparent since the application of forward pressure by the DUT/SUT enables interfaces to relieve saturated output queues by forcing access to the medium and concomitantly inhibiting the tester from transmitting frames. I have added to the See Also accordingly: See Also: See also: intended load (3.5.1) [ADDED] offered load (3.5.2) [ADDED] overloading (3.5.4) [ADDED] forwarding rate (3.6.1) [ADDED] backpressure (3.7.1) > --------------------- 3.7.3 Head of line blocking Definition: Frame loss observed on an uncongested output interface whenever frames are received from an input interface which is also attempting to forward frames to a congested output interface. -- not just loss, also added delay <Bob replies: yes, I had that in there at one point but took it out for no good reason. So back in it goes. Actually it is in the discussion already. Definition: Frame loss or added delay observed on an uncongested output interface whenever frames are received from an input interface which is also attempting to forward frames to a congested output interface. > --------------------- 3.10.1 Broadcast forwarding rate Definition: The number of broadcast frames per second that a DUT/SUT can be observed to deliver to all interfaces located within a broadcast domain in response to a specified offered load. -- I'd tweak The number of broadcast frames per second that a DUT/SUT can be observed to deliver to all interfaces located within a broadcast domain in response to a specified offered load of frames to the broadcast MAC address. <Bob replies: done with "destined to" added: Definition: The number of broadcast frames per second that a DUT/SUT can be observed to deliver to all interfaces located within a broadcast domain in response to a specified offered load of frames directed to the broadcast MAC address. > what about multicast? <Bob replies: I don't want to constrain Kevin Dubray in his work on multicasts by offering a preemptive definition. Multicasts are complex entities and deserve a document of their own.> ----------------------- 4. Security Considerations This document raises no security issues. -- the iesg will no longer OK RFCs that have this type of Security Considerations section - it need to say things like Documents of this type do not directly effect the security of the Internet or of corporate networks as long as testers implementing tests of this type are not connected to operating networks. The document points out that switching devices may violate the IEEE 802.3 standard by transmitting frames below the minimum interframe gap or unfairly accessing the medium by inhibiting the backoff algorithm. Although such violations do not directly engender breaches in security, they may perturb the normal functioning other interworking devices by obstructing their access to the medium. <Bob replies: done with your wording plus an additional paragraph on standards violations. - thanks for the tip. Documents of this type do not directly effect the security of the Internet or of corporate networks as long as benchmarking is not performed on devices or systems connected to operating networks. The document points out that switching devices may violate the IEEE 802.3 standard by transmitting frames below the minimum interframe gap or unfairly accessing the medium by inhibiting the backoff algorithm. Although such violations do not directly engender breaches in security, they may perturb the normal functioning of other interworking devices by obstructing their access to the medium. Their use on the Internet or on corporate networks should be discouraged. > Scott <Bob replies: Thanks for the broad range of your comments Scott, they cover formal points, some very fine points and some minor wording issues, all of which had to be addressed. Hope to see this work go forward now and to catch up with you in Munich.> Bob
- Switch draft Q&A Kevin Dubray