If memory serves me right, the first set of tests were not using "ip
flow-cache feature-accelerate". The 2nd set (which reached 45K PPS for
minimal sized packets) did use this, and performamnce indeed accelerated.
You need also to note that some effort was invested in distributing the
load between several CPUs on each router (the main RSP4 and two VIPs). At
45K PPS and an average packet size of 500 bytes, you are in the OC3 ball
park.
Oded.
On Tue, 11 Jan 2000, Hank Nussbacher wrote:
>>Resent-Date: Tue, 11 Jan 2000 11:41:41 -0500
>>Date: Tue, 11 Jan 2000 10:39:06 -0600
>>From: Edward Henigin <ed@staff.texas.net>
>>To: Cisco NSP <cisco-nsp@puck.nether.net>
>>Subject: Re: policy routing performance
>>X-Mailer: Mutt 0.93.2i
>>Resent-From:cisco-nsp@puck.nether.net
>>X-Mailing-List: <cisco-nsp@puck.nether.net> archive/latest/2005
>>X-Loop: cisco-nsp@puck.nether.net
>>Resent-Sender: cisco-nsp-request@puck.nether.net
>>
>>
>> Available in the 'T' train, so if I want that and turboacl
>>I will see them together in 12.1?
>>
>> Hank: were those tests done using NPR? Based on your config
>>(since I saw 'ip route-cache policy' AND 'ip route-cache flow' both
>>under the interface) I would guess that yes, that included NPR.
>>If so, the performance seems as though it leaves a bit to be desired.
>>It looks as though it uses 15% cpu per 3Kpps, it looks as though
>>you'd be pegged at around 18Kpps. (I'm going to need scalability
>>beyond that...) Then again, that was through a tunnel, and I do
>>believe that going through a tunnel increases the overhead.
>>
>> Ed
>>
>>--
>>On Tue, Jan 11, 2000 at 07:08:05AM -0800, Barry Raveendran Greene said:
>>> Hello Ed,
>>>
>>> Try Netflow Policy Routing:
>>>
>>>
>http://www.cisco.com/univercd/cc/td/doc/product/software/ios120/120newft/120
>>> t/120t3/policyrt.htm
>>>
>>> Barry
>>
>>
>
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:09 EDT