4

Does it make more sense to modify the 1800 second refresh rate, pacing timers, or all of the above in terms of network stability/best practices/etc to achieve quicker LSA maintenance?

I am dealing with a specific situation wherein modifying the default values is desired. What are some of the typical adjustments made away from the defaults in a conservative (not too agressive) way?

  • 1
    Sometimes it is worth modifying OSPF defaults... we need to know a lot more details about the problem you are really solving, otherwise you will get rampant hand-waving and speculation. – Mike Pennington May 31 '13 at 2:28
  • This is in reference to R&D testing for a technological device that involves OSPF. There are very particular test cases defined by a 3rd party that would require changing the default refresh rate to more quickly adapt to forced breakage of segments in the network that we know will occur during certain tests. – THE DOCTOR Jun 3 '13 at 21:35
  • Did any answer help you? if so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer. – Ron Maupin Aug 8 '17 at 9:36
2

Honestly if you're concerned with best practices and keeping your LSDBs nice and neat, ready for fast convergence, I would recommend ensuring that your OSPF areas are constructed well. Keep LSAs that don't need to be in an area summarized, and don't let a single area grow too large.

| improve this answer | |
  • This makes sense in enterprise, but not in ISP – mellowd May 31 '13 at 7:06
  • Why not? Just curious, I have never worked in ISP space but I have heard this before. – Mierdin Jun 2 '13 at 17:02
  • 1
    The primary reason is for MPLS traffic engineering, Type10 LSAs, which include link affinities, have area-wide scope. If you divide your network into smaller areas, you lose the ability to use that information – mellowd Jun 2 '13 at 17:09
  • See my comment below against Adam Loveless's answer, I think the same point applies here: controlling the flooding and pacing of LSAs across the IGP domain can make a difference to control-plane utilisation. If using MPLS-TE for example as @mellowd said, one can end up reducing information across the IGP domain which could be bad, so pacing and packaging the information distribution to make it more efficient can be beneficial. Some Cisco IOS examples: null.53bits.co.uk/… – jwbensley Apr 16 '17 at 10:02
1

My opinion is that you leave the timers as default and use features like BFD to trigger and thus shorten re-convergence times.

Can you give some more detail around what issue you are trying to solve modifying the refresh timers? Hello and Dead timers are what you'd tweak if you're looking for faster convergence time.

The Cisco recommendation above about 250msec and 1 second above is only valid if you have point-to-point ethernet links everywhere, and can guarantee that your area will always be this way.

In an OSPF broadcast network, the last thing I'd want is an RSTP/MSTP/RPVST+ re-convergence (1-2 seconds) to trigger an OSPF re-calculation (750msec-1 second)

| improve this answer | |
0

90% of the time it makes sense to leave the defaults as-is. But in a high availability environment or large ISP it does make sense to adjust the timers to exactly what you need

| improve this answer | |
0

What are some of the typical adjustments made away from the defaults in a conservative (not too agressive) way?

It is recommended that the OSPF hello and dead timers be reduced to 250 msec and 1 second, respectively.

Source

| improve this answer | |
  • I would rather leave hellos to defaults and enable BFD – mellowd May 30 '13 at 18:42
  • +1 - Reducing hello timers on IGP's isn't a great way to go. BFD is, by far, a much more stable (read: better) answer. – rnxrx May 31 '13 at 15:29
  • Reducing the hello timers and/or adding in BFD is simply to detect link/interface failures rapidly. This doesn't help with, for example, an interface coming up and a huge flood of LSAs being advertised to a neighbor. There are many options to tune beyond just hello timers. This can have an impact on the control-plane in large IGP domains. – jwbensley Apr 16 '17 at 9:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.