OSPF maximum number of adjacencies?

phil_carter · ‎08-25-2011

Hi,

Running around 100 OSPF adjacencies on a single router CISCO2911/K9, and intend to add a further 100-200 on top of this. Is there a restriction to the number permitted/recommended? The router is acting as a single core node in OSPF Area 10, with uplinks to Area 0. All remote sites beyond this are via ADSL and run no OSPF summarisation etc - just standard OSPF neighbours in Area 10 with all routes known on every router.

MRCW_R29_496#sh ip ospf statistics

OSPF Router with ID (10.200.1.30) (Process ID 1)

Area 0: SPF algorithm executed 369 times

Area 10: SPF algorithm executed 209270 times

Summary OSPF SPF statistic

SPF calculation time
Delta T   Intra D-Intra Summ    D-Summ Ext     D-Ext   Total   Reason
00:04:57   4    0       20      12      68      4       112     R
00:04:47   4    4       20      12      68      60      168     R, X
00:03:47   4    0       20      12      68      4       112     R
00:03:37   4    4       20      16      64      60      168     R, X
00:02:37   4    0       20      12      68      4       112     R
00:02:27   4    4       20      16      64      60      168     R, X
00:01:27   4    0       20      12      68      4       112     R
00:01:17   4    4       20      16      64      60      168     R, X
00:00:17   4    0       20      12      68      4       112     R
00:00:07   4    4       20      16      64      60      168     R, X

MRCW_R29_496#sh mem
                Head    Total(b)     Used(b)     Free(b)   Lowest(b) Largest(b)
Processor   28F4D520   344664800    32102944   312561856    76229296   245631188
      I/O    D800000    41943040    19064332    22878708    22583248    21746300

IOS: c2900-universalk9-mz.SPA.150-1.M4.bin

As the above shows, there's plenty of free memory on the core router, but I have seen problems on 6500s in the past where once a certain level of adjacencies is created, duplicating routing updates caused an OSPF meltdown. OSPF single point of failure aside, are their concerns for the large addition of adjacencies planned?

rgds

Phil

Amit Aneja · ‎08-25-2011

Phil,

Can't give you a hard limit, but, I guess upto you shouldn't really have problem.

For 300, If you are having that many neighbors on a 2900 you probably have a design problem. This is just my personal opinion.

Regards,

Amit

Peter Paluch · ‎08-25-2011

Amit,

Personally, I also believe that while technically possible, it is somewhat ... unusual to keep so many neighbors to a single router, to say the least.

One thing I would have a look onto would be the CPU load. The need to generate and process hundreds of Hellos every 10 seconds, maintain hundreds of LSA-1 in the LSDB, process them with each their update, receive their reoriginated contents roughly every 30 minutes - a CPU load profiling would be nice here. Phil, if possible, do try to make a profile of your CPU load throughout a day. You may do a crude estimation that with twice so many OSPF neighbor, the CPU load may also rise twice as much. Also try to see how the CPU spikes up if you add or remove a directly connected network on a router (a loopback is sufficient) or change some router's ID, as that should result in running full SPF.

Best regards,

Peter

Jon Marshall · ‎08-25-2011

Phil

It also depends on how stable the network is. If you have links going up and down often this will affect the router quite a bit. ADSL links from my experience are not the most stable of links so link changes would generate LSA's that need to be propogated throughout the OSPF area.

It's basically a suck it and see approach unfortunately as there are no hard and fast rules when it comes to number of adjacencies supported.

Jon

phil_carter · ‎08-26-2011

Thanks for the replies - as I expected - no real set in stone answer. However, looking at the number of full SPF recalculations, there are a lot on a daily basis:

MRCW_R29_496#sh ip ospf statis

OSPF Router with ID (10.200.1.30) (Process ID 1)

Area 0: SPF algorithm executed 369 times

Area 10: SPF algorithm executed 211332 times

Summary OSPF SPF statistic

SPF calculation time
Delta T   Intra D-Intra Summ    D-Summ Ext     D-Ext   Total   Reason
00:05:27   4    0       20      12      68      8       116     R
00:05:17   4    4       20      16      68      60      172     R, X
00:04:17   4    0       20      16      68      4       116     R
00:04:07   4    4       20      16      64      60      168     R, X
00:03:07   4    0       20      12      68      4       112     R
00:02:57   4    4       20      16      68      60      172     R, X
00:01:57   4    0       20      12      68      4       112     R
00:01:47   4    4       20      16      64      60      168     R, X
00:00:47   4    0       20      12      68      4       112     R
00:00:37   4    4       20      16      64      60      168     R, X

MRCW_R29_496 uptime is 22 weeks, 6 days, 18 hours, 50 minutes

...thats around 1300 SPF recalculations a day... this seems fairly high? Does this sound excessive?

Monitoring the CPU today, so far it hasn't gone much above the 4% mark.

boss.silva · ‎08-26-2011

Hello,

The limit will be your hardware resources

As long as you still have CPU and memory, it'll work.

But just so we understand, would it be possible to show us what kind of design you have there? We are curious to know what situation would have that many adjacencies to a single router.

Regards,

Bruno Silva.