[j-nsp] M10 FEB heap usage (RPF/route options)

Kevin Day toasty at dragondata.com
Fri Dec 14 18:14:14 EST 2007


On Dec 14, 2007, at 5:02 PM, Nicolaj Kamensek wrote:
>>   Heap utilization                   97 percent
>
> what does
>
> show arp no-resolve | count
>
> say?
>

admin at core-ams> show arp no-resolve |count
Count: 411 lines

>> Is this normal? I seem to remember being able to fit 500k+ routes  
>> not  that long ago (multiple ribs). Did the size of each v4 route  
>> entry  grow during a junos upgrade at some point?
>
> 500k+ routes is still possible. Those routes are stored in the 8MB  
> SSRAM on your FEB but also a copy of this table is stored in the  
> DRAM, 64MB in your case. You reached the limits of your DRAM, not  
> the SSRAM. Upgrading it should help, either by buying a newer router  
> like M7i/M10i(with 256MB DRAM on the FEB) or by installing a  
> enhanced FEB for M10 routers.
>
>> Does this mean those of us stuck on 64MB FEB/SSB systems are going  
>> to  have to run without RPF now or in the very near future when the  
>> size  of a full table grows a bit more?
>
> Pretty much, yes. 64MB isn't simply large enough for todays full- 
> table + arp entries + RPF + ... 128MB or more might be sufficient  
> for that.
>
> We reached our limits on a M20 router with SSB-E with about 7500 arp  
> entries, full-table and no RPF. Upgrading to SSB-E16 solved that  
> problem.
>

What I meant to say is that I seem to remember that we were able to do  
500K routes on the same hardware with the same features enabled, in  
previous versions of JUNOS. I've been told that you were able to do  
700-800k routes in an IP2 in the 4.x days. I don't think that's  
possible in 8.x no matter how much you turn off.

It would be really nice for Juniper to publish a list of how much  
memory per route/feature is needed in which versions of JUNOS - I  
think a bunch of people are going to be hitting limits like this  
unexpectedly after an upgrade, especially with the recent security  
issue causing some to upgrade sooner.

-- Kevin



More information about the juniper-nsp mailing list