[c-nsp] Memory utilization threshold

[Ihsan Junaidi Ibrahim]

On Thu, 13 Oct 2005 11:31:53 +0800, Rodney Dunn <rodunn at cisco.com> wrote:

> If you have dual paths and a full BGP feed 256M is not enough
> anymore due to the BGP table growth and IOS image size increases.
>
> And the transient memory usage has been improved in newer code
> (ie: late 12.0S).
>
>
> On Thu, Oct 13, 2005 at 11:32:24AM +0800, Ihsan Junaidi Ibrahim wrote:
>> On Wed, 12 Oct 2005 21:48:13 +0800, Rodney Dunn <rodunn at cisco.com>  
>> wrote:
>>
>> > I've always said the 60-70% number for CPU is pretty reasonable
>> > given my experience troubleshooting broken networks.
>> >
>> > As for memory, that's a much harder question to answer because
>> > it depends on the deployment. For full BGP feed routers I'd
>> > recommend alarms at 40M free and get really concerned at 20M
>> > free. It's due to the use of transient memory for reconvergence.
>> > That doesn't allow for any room to grow other things on the platform
>> > like # interfaces, features, etc...
>> >
>> > As for a router not doing large routing tables you threshold can
>> > be much smaller depending on the feature set.
>> >
>> > Maybe a better approach is I'm a big fan of customers trending
>> > their network and then set a threshold for alarm at some percentage
>> > off the average.
>> >
>> > Rodney
>> >
>>
>> Can this be applied on an "old design" 7507 with 256 MB RAM on an RSP4?
>> Additionally this router serves as one of our primary border routers so  
>> is
>> there any other justfication for at least 40M reserve memory as an alarm
>> threshold?
>>
>> --
>> Thank you for your time,
>> Ihsan Junaidi Ibrahim

Is there a technical documentation or white paper that highlights this  
very issue? At the moment we are not experiencing any significant problems  
with regards to the utilization but we want to be in the position where we  
can justify the upgrade to the non-tech decision makers should the need  
arises.

-- 
Thank you for your time,
Ihsan Junaidi Ibrahim