Will there be a difference in CPU due to OSPF in same AREA

arun kumar · ‎09-18-2008

Hi,

We have around 300 routers in our area 0 in OSPF. But we are observing high CPU due to OSPF process only on some routers not on any other routers. I would like to know if any SPF calculation happens that update will be propagated to all the routers in the same area and every router will have a same impact (i mean the routers with same platform, memory, IOS and load). But it is not happening uniformly in our network. Is there any specific thing that I need to check.

Thanks in advance.

regards,

Arun Kumar

Giuseppe Larosa · ‎09-18-2008

Hello Arun,

even if all routers are the same platforms with same cpu, memory and IOS image and are in the same area there are some possible reasons for an higher CPU load on some of them :

a) the OSPF DR role implies more work: this is specially true if the same router is OSPF DR on multiple lan segments: ip ospf priority can be used to distribute the DR role on a per interface basis

b) additional duties like ABR (router working as point of contact between different areas it has a db for each area) and ASBR (router importing routes in the OSPF domain by use of some form of redistribution)

c) different SPF timers settings that can make a router to execute the SPF more often then others and so it can experience higher CPU usage caused by OSPF

this can be checked with show ip ospf

A final note is that 300 routers could make one to think to use multi-area (if not already in use).

Hope to help

Giuseppe

Joseph W. Doherty · ‎09-18-2008

Besides the good information Giuseppe provided, I have a hazy recollection that some of the newer Cisco OSPF implementations might support partial SPF calculation. I.e. instead of rebuilding the whole SPF tree, they only recaculate, perhaps, a branch.

If this is true, depending what has changed within the area topology and where a particular router is within the topology, the impact of SPF recalculation could vary.

Joseph W. Doherty · ‎09-18-2008

Found the feature that I had the hazy recollection on; it's called "incremental SPF". More information can be found on it here http://www.cisco.com/en/US/docs/ios/12_0s/feature/guide/ospfispf.html but it appears to be an option that needs to be enabled. So unless you have this option enabled on your platforms, I doubt this is the reason for your CPU differences.

PS:

Perhaps another reason for different CPU impact, depending on how SPF is implemented, the SPF tree often will appear different to each router, its appearance depends on that router's position in the topology.

arun kumar · ‎09-19-2008

Hi Joseph,

We already enabled ISPF on those routers but still getting high CPU which is also showing OSPF as the one of the top contributor.

regards,

Arun

Joseph W. Doherty · ‎09-19-2008

If you already have enabled ISPF, then it might account for some of why your routers that are otherwise identical are seeing different CPU loads attributed to OSPF. Also, ISPF doesn't guarantee you still won't see high CPU due to OSPF, only that it might reduce the impact.

If your concern is the high CPU load due to OSPF in general, then Giuseppe's concern about having 300 routers in one area, and perhaps placing them in multi-areas, should be further considered.

ccollister · ‎09-19-2008

300 definitely seems high--isn't 50 the preferred max per area?

Joseph W. Doherty · ‎09-19-2008

I've also seen 50 as a recommendation for max routers per area, but you really need to account for how complex the topology is and how powerful the routers are. As with all the other posters, 300 does seem a might many.

m-haddad · ‎09-19-2008

Hello,

Having 300 routers in one area will cause SPF calculation to be done by all routers in the area when routes change. This design is not considered according to best practices. Routes summarization and using multi-area design could lower the impact/occurency of SPF calculation on routers.

Hope this helps,

Regards,