[26801] in North American Network Operators' Group
RE: CEF Load balancing...
daemon@ATHENA.MIT.EDU (Martin, Christian)
Mon Jan 17 10:58:46 2000
Message-ID: <4F8C08CC6E76D311AD1F00508B78724907E9E9@HERMES>
From: "Martin, Christian" <CMartin@mercury.balink.com>
To: "'Forrest W. Christian'" <forrestc@iMach.com>
Cc: "'nanog@merit.edu'" <nanog@merit.edu>
Date: Mon, 17 Jan 2000 10:51:46 -0500
MIME-Version: 1.0
Content-Type: text/plain;
charset="windows-1252"
Errors-To: owner-nanog-outgoing@merit.edu
>We have a situation where CEF in a particilar 12.0.x version
>broke the BVI
>interface in an IRB configuration w/ATM. Then the next version broke
>forwarding between an ATM interface and a Ethernet interface in an
>interesting way (I haven't tracked down the specifics on the
>second case
>yet.)
I've seen issues where CEF drives the CPU through the roof if it is doing
per-packet balancing on interfaces that reside on different VIPs. I've also
seen links drop out of a CEF load-balancing group and the Adj Table hashes
not purge the down interface, or keep the hash indices the same, with only
two interfaces instead of three, thereby doing 67/33 balancing instead of
50/50. The workaround is to diable and reenable CEF.
On the positive side, you can easily aggregate 200+Mbps across 6 DS3s with
minimal CPU and smooth, per-packet balancing. Process switching on steroids
it is... Besides, to uses reverse-path filtering, dCAR, dWRED, etc, you
need dCEF.
It is stable in 11.1.26CC1 - we use it all over for load-balancing on
DS3s...
Chris
>
>All in all, CEF works wonders, if it doesn't break something with its
>extra-fancy forwarding.
>
>- Forrest W. Christian (forrestc@imach.com) KD7EHZ
>----------------------------------------------------------------------
>iMach, Ltd., P.O. Box 5749, Helena, MT 59604 http://www.imach.com
>Solutions for your high-tech problems. (406)-442-6648
>----------------------------------------------------------------------
>
>
