From: Mohammed Alwosh Al-Samarrai (malwosh@xxxxxxxxx)
Date: Wed Jun 13 2001 - 02:58:17 GMT-3
Ok guys.
Out of order packets posses an issue specially with the fact that you can
fill the buffers of the end-receiving destination and thus can cause the
machine to crash because it simply can't handle all these packets while it
is still waiting for the later-sequence packets!!
Now, what you can do inside your Cloud is to change the MTU size and chop
those large packets into smaller packets. This way, provided that you are
using the suitable Queuing and Quality of Service techniques, you prevent
tail-drops from occurring. The chopped packets are layer 3 packets with each
one having the Layer 3 src and destination address and the M(more) bit set
until the last packet of the flow.
This is what I know. Hope it helps.
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
Mas Kato
Sent: 13 June 2001 00:19
To: 'Chuck Church'; 'CCIELAB'
Subject: RE: WFQ and BRIs
Indeed, it'd be interesting to hear from the guys and gals working in
ultra-high-performance environments. One would think Netflow and/or dCEF
could functionally offer something like "per-socket" load-balancing...
Mas
-----Original Message-----
From: Chuck Church [mailto:cchurch@MAGNACOM.com]
Sent: Tuesday, June 12, 2001 11:51 AM
To: Mas Kato; 'CCIELAB'
Subject: RE: WFQ and BRIs
Good article. Brings up a question, at least to me anyway. Since
packet
reordering is so bad, is it the general consensus of the ISP world to
load
balance based on per-destination rather than per-packet? It's just
something that I've been wondering about for some time.
Thanks,
Chuck
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
Mas Kato
Sent: Tuesday, June 12, 2001 2:15 PM
To: 'Rick Stephens'; 'CCIELAB'
Subject: RE: WFQ and BRIs
Ah! Just like a heavily loaded Juniper box! <G>
http://www.lightreading.com/document.asp?doc_id=4009&page_number=8
Thanks Rick!
-----Original Message-----
From: Rick Stephens [mailto:rstephens@wantec.com]
Sent: Tuesday, June 12, 2001 10:38 AM
To: 'Mas Kato'; 'CCIELAB'
Subject: RE: WFQ and BRIs
Here is what I turned up. This is related to bridging, but it appears to
be
much more widely disabled in standard configurations. Probably for
similar
issues.
CSCdm45164
Enabling weighted fair queuing (WFQ) on an interface that belongs to a
transparent bridging bridge group may cause packets that are egressing
that
interface to be sent out of order. This situation causes failure in
terminated and bridged Logical Link Control 2 (LLC2) sessions.
Workaround: Disable WFQ using the no fair-queue interface configuration
command.
-----Original Message-----
From: Mas Kato [mailto:tealp729@home.com]
Sent: Monday, June 11, 2001 1:21 PM
To: 'CCIELAB'
Subject: WFQ and BRIs
Many moons ago, turning off weighted fair queueing ('no fair-queue') on
BRIs was standard practice. I don't remember why. Was it a workaround
for a code-path problem or something? An archive search turned up zilch.
Anyone?
Mas
**Please read:http://www.groupstudy.com/list/posting.html
**Please read:http://www.groupstudy.com/list/posting.html
**Please read:http://www.groupstudy.com/list/posting.html
**Please read:http://www.groupstudy.com/list/posting.html
This archive was generated by hypermail 2.1.4 : Thu Jun 13 2002 - 10:31:23 GMT-3