I dropped out of the telecom last week for a while and I missed that
discussion. From my point of view, small message latency is the most
important figure of merit.

I agree with Randy that there are two typical message types, very large and
very small. Many applications consist of a data plane and a control plane.
The data plane typically has predictable flows and needs high bandwidth, but
not low latency.

The control plane uses small messages (10-20 bytes). (You can get a feel for
it by estimating the amount of data in a typical function call. How often do
you pass 1KB in parameters?) The message traffic is sporadic and bursty.
Latency is critical here because (1) the messages are generally smaller than
the product of bandwidth and latency (1Gb/s * 1us = 125 bytes) and (2) you
can't use pipelining to speed up a critical path that sends a message and
waits for a response.

Dominic Herity
www.redplain.com
 

-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of
[EMAIL PROTECTED]
Sent: 02 June 2007 20:13
To: [email protected]
Subject: tipc-discussion Digest, Vol 13, Issue 1

Send tipc-discussion mailing list submissions to
        [email protected]

To subscribe or unsubscribe via the World Wide Web, visit
        https://lists.sourceforge.net/lists/listinfo/tipc-discussion
or, via email, send a message with subject or body 'help' to
        [EMAIL PROTECTED]

You can reach the person managing the list at
        [EMAIL PROTECTED]

When replying, please edit your Subject line so it is more specific
than "Re: Contents of tipc-discussion digest..."


Today's Topics:

   1. TIPC WG: Minutes of May 31, 2007 Meeting (Stephens, Allan)


----------------------------------------------------------------------

Message: 1
Date: Fri, 1 Jun 2007 13:35:36 -0700
>From: "Stephens, Allan" <[EMAIL PROTECTED]>
Subject: [tipc-discussion] TIPC WG: Minutes of May 31, 2007 Meeting
To: <[email protected]>
Message-ID:
        
<[EMAIL PROTECTED]>
Content-Type: text/plain;       charset="us-ascii"

A phone meeting of the TIPC working group was held on May 31, 2007.
 
Attendees:
Al Stephens (Wind River) -- meeting chair; Jon Maloy (Ericsson); Randy
MacLeod (Nortel); Dominic Herity (Red Plain); Elmer Horvath (Wind River)

Minutes:
 
1. Meeting opening

The meeting was called to order at 11:03 AM (EDT).
 
2. TIPC BOF at Ottawa Linux Symposium

The agenda for the TIPC birds-of-a-feather session at OLS was agreed
upon.  Randy will begin with a short presentation about TIPC and his
experiences as a user; Al will provide an update on development
activities over the last year and outline the current roadmap for future
development; Dominic will talk about Red Plain's experience in porting
their distributed object product from IP to TIPC.  All presentations
will be kept short to allow significant time for questions and
discussion with session attendees.

ACTION: All presenters to distribute their slides for comment by June
15th (send to meeting attendees only).

It was also agreed that the TIPC Working Group will not hold a June 28th
conference call, as we will be meeting at the BOF around that time
anyway.

3. TIPC Roadmap

Al asked participants about their highest priorities for upcoming
development of TIPC.

Randy:
1) Control/data plane separation (as mentioned in previous emails); may
be needed in Fall 2007 timeframe.
2) Improved support for non-continguous sk_buffs.  This covers both the
egress side of things (eg. the ability to create large messages using
multiple chunks rather than a single large chunk, which may be
unavailable due to memory fragmentation issues) and the ingress side of
things (eg. the ability to handle non-linear messages and/or fragmented
messages without having to copy their component parts into a single
large chunk).

Randy also stated that he was not aware of a near-term need for
hierarchical TIPC by Nortel.

Jon:
1) Support for up to 64 (or even 256) bearers to allow users to
configure a distinct VLAN-based bearer for each neighbouring node in a
cluster.  [Note: Al observed that this can be done right now with a
trival change to TIPC_MAX_BEARERS, but Jon would like to avoid
allocating memory for these bearers until they are actually needed.]
2) Support for link subscriptions.  [Note: This might be necessary to
support the control/data plane separation Randy wants to achieve.]

Jon also thinks that creating a TCP-based bearer for TIPC (possibly
running directly over Ethernet, rather than over IP) might be desirable
performance/reliability reasons.

Al will discuss this info with his superiors and put together a
tentative TIPC roadmap.

4. Mailing list issues

Randy raised concerns about the stability of SourceForge, observing that
the tipc-discussion mailing list has become somewhat unreliable at
times, and downloads/web pages are sometimes inaccessible.  Al suggested
we discuss our concerns with other project developers at OLS, and see if
they can suggest better alternatives (such as code.google.com).

5. TIPC message sizes

Elmer is interested in knowing what "typical" message sizes are for TIPC
users, as this info will allow the development team to concentrate
performance optimization efforts on the most commonly used areas.  Randy
observed that applications he has encountered typically involve small
numbers of small packets (i.e. 10's of bytes) or large numbers of large
packets (i.e. 1000's of messages, each up to 50KB); in both cases, the
messages were being sent in a connectionless manner.  In the former case
his expectation is that TIPC should provide excellent performance vs.
UDP, while in the latter he would expect good performance (i.e.
comparable, or better) vs. UDP.

6. Meeting close

The meeting adjourned at 11:40.  The next meeting will be at OLS, as
mentioned previously.
 
Please let me know of any errors or omissions.
 
Regards,
Al Stephens
 
[END]



------------------------------

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

------------------------------

_______________________________________________
tipc-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/tipc-discussion


End of tipc-discussion Digest, Vol 13, Issue 1
**********************************************





-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
tipc-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/tipc-discussion

Reply via email to