Fw: [c-nsp] A Stable BRAS IOS? + performance

[Osama I. Dosary]

Thanks Gyebnár,
But we are using a keepalive of 60 seconds!
Could it be the fragmentation that is causing low performance?
How can I tell?
"sh ip traffic" gives absolute values. What is a high fragmentation rate 
or percentage?
#sh ip traffic Output:
IP statistics:
  Rcvd:  206704290 total, 58613783 local destination
         7947 format errors, 7 checksum errors, 79 bad hop count
         0 unknown protocol, 35440 not a gateway
         0 security failures, 0 bad options, 12419 with options
  Opts:  0 end, 0 nop, 0 basic security, 0 loose source route
         0 timestamp, 0 extended security, 0 record route
         0 stream ID, 0 strict source route, 12419 alert, 0 cipso, 0 ump
         0 other
  Frags: 276689 reassembled, 575 timeouts, 0 couldn't reassemble
         9383185 fragmented, 0 couldn't fragment

Gyebnár Krisztián wrote:

>
> Hi,
>
> We running on 12.3.10a on our NPE-G1 without any problem, the 
> performance:
> ~3000 users, no fragmantation(fully 1552 byte MTU path from the LAC to 
> LNS)
> 250Mbit/sec bidirectional taraffic and the cpu about 55-60%, also only 
> L2TP
> and PPP session termination is going on.
>
> Important: check the keepalive time because the default is 10 seconds 
> :-(,
> this cause many overhead on the Network, CPU...
> we set up this to 30s and the performance is much better :-)
>
> Krisztián
>
>> ----- Original Message ----- From: "Osama I. Dosary" <oid at saudi.net.sa>
>> To: <cisco-nsp at puck.nether.net>
>> Sent: Monday, December 06, 2004 7:50 AM
>> Subject: Re: [c-nsp] A Stable BRAS IOS? + performance
>>
>>
>>> Robert E.Seastrom wrote:
>>>
>>>> "Osama I. Dosary" <oid at saudi.net.sa> writes:
>>>>
>>>>
>>>>> Hello,
>>>>>
>>>>> We are using Cisco routers (NPE-G1) as a NAS and LNS. No PVCs are
>>>>> terminated on these routers, only L2TP and PPP session termination is
>>>>> going on.
>>>>>
>>>>> We have having difficulting finding a stable IOS to run on these
>>>>> routers.
>>>>>
>>>>> Does anyone know of a good stable IOS they have been using for the
>>>>> same/similar purpose, and recommend?
>>>>>
>>>>
>>>> I've been using c7200-js-mz.123-8.T4.bin for about six weeks now as an
>>>> LNS; while it's not GD it runs quite nicely on an NPE300 and I believe
>>>> will run on your G1 as well.  Our load peaks at a little over 1600
>>>> simultaneous users, 55% CPU, 50 Mbit aggregate (up + down) traffic
>>>> through router (your mileage will obviously vary since you have a much
>>>> more studly CPU).  CPU load is slightly inflated because of
>>>> outstanding issues with a few hosts that are moving oversized L2TP
>>>> packets that get fragmented and need reassembly to pop the PPPoE frame
>>>> (fragment reassembly is most assuredly *not* in the fast switching
>>>> path).
>>>>
>>>>
>>> On our NPE-G1 (for Dialup) load peaks at ~3000 users, 110Mbps (up+down)
>>> traffic and CPU util is about 80%. I feel like we are doing something
>>> wrong, according to Cisco docs, this router should be able to handle 
>>> tons
>>> more.
>>>
>>> Even when we used to run NPE-300s for dialup, it could not handle more
>>> than 900 sessions, with CPU over 90%. What could be the reason?
>>> We used various IOS versions, without much difference in 
>>> performance. We
>>> noticed that when the B-train version is used we save about 5-10% on 
>>> CPU
>>> cycles. But it is still far from the router specifications.
>>>
>>> Could this be a matter of fragmentation of the L2TP packets that 
>>> could be
>>> causing bad performance.
>>> Any ideas?
>>>
>>>
>>> -Osama
>>>
>>>
>>>
>>> _______________________________________________
>>> cisco-nsp mailing list  cisco-nsp at puck.nether.net
>>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>>> archive at http://puck.nether.net/pipermail/cisco-nsp/
>>>
>>
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>