Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[8023-CMSG] [Fwd: FW: [802.1] Per-priority Pause]




Norm,

Matt sent me this email from the 802.1 reflector in response to
my 802.3 Congestion Management reflector request regarding
how well Muneyoshi Suzuki's PAUSE per PRIORITY presentation
was received within 802.1. Thanks for the reasoned explanation
behind the comments at the 802.1 meeting.

You mention "Placing the buffers in the proper device". I'll presume
this means the NIC/end stations, the things some say don't typically
support priorities. I think you've mentioned before that there's
nothing in 802.1 that prohibits NICs from supporting priorities.
Could you please confirm this?

To pursue this line further, you mention:

> Any solution to this problem that involves the PC recognizing that
> traffic to its printer is a different flow than traffic to the
> network can be trivially solved by telling the PC to limit its flow
> of traffic to the network to the contracted 20 Mbits.

This sounds much like some of the work we're proposing in
congestion management, a means of limiting the data rate of
a link. However, there is a hint of another idea here that I'd
like you to comment on.

Printer traffic is a different flow than traffic to the network,
thus it seems logical that since you are also suggesting that
the PC add priority queue support to its NIC, that this traffic
should wind up in different priority queues. Now, when you
suggest that the network traffic should be limited to 20 Mb/s,
are you suggesting there may be advantages to being able to
limit the data rate on a per-priority queue basis?

Perhaps I've let my mind wander a bit too far this morning but
I'm curious to hear your comments (as well as those of other
folks).

Thanks,
Ben

--
-----------------------------------------
Benjamin Brown
178 Bear Hill Road
Chichester, NH 03258
603-491-0296 - Cell
603-798-4115 - Office
benjamin-dot-brown-at-ieee-dot-org
(Will this cut down on my spam???)
-----------------------------------------



fyi	

> -----Original Message-----
> From: Norman Finn [mailto:nfinn@cisco.com]
> Sent: Thursday, July 15, 2004 5:16 PM
> To: STDS-802-1-L@listserv.ieee.org
> Subject: [802.1] Per-priority Pause
> 
> 
> Muneyoshi,
> 
> On further consideration of your per-priority Pause proposal, 
> I can see
> better the very good reasons why you proposed it, and would like to
> explain the rather violent negative reaction that your 
> proposal received
> from some 802.1 and 802.3 people.
> 
> When intelligent, competent people disagree strongly, the reason is
> almost always due to differing assumptions.  I think I've found the
> different assumptions that we were making.
> 
> For the particular diagram you were showing -- a telephone and a PC
> connected via a 100Mb repeater to a small media converter -- 
> a per-port
> Pause might be a cost-effective solution to a real problem.  This is,
> of course, why you brought the idea into 802.1.
> 
> The concern that I and others expressed was based on a different set
> of assumptions, namely, that 802.3 is a network medium, not an access
> medium.  In a network medium, per-anything Pause raises all of the
> unsolved problems that were mentioned in the meeting.  Per-anything
> Pause is simply a very bad idea for an IEEE 802.1/802.1 network, and
> probably, for any 802 network at all.
> 
> This can best be explained by very slightly altering your diagram.
> Just add a printer to the repeater.  Now, the per-priority Pause
> frames will stop the PC-printer traffic, as well as the PC-network
> traffic.
> 
> Any solution to this problem that involves the PC recognizing that
> traffic to its printer is a different flow than traffic to the
> network can be trivially solved by telling the PC to limit its flow
> of traffic to the network to the contracted 20 Mbits.
> 
> Placing the buffers in the proper device, according to the present
> 802.1 and 802.3 standards solve the problem very well.  Perhaps that
> is your best bet for solving this access problem.
> 
> -- Norm
> 
> IEEE 802.1 list: see 
http://www.ieee802.org/1/email-pages/tbyoc604.html