Re: OSPF Performance

Hi,

In general, throttling timers should be tuned according to your goals.
Often, the primary goal is fast network convergence. In this case, you
should be mainly concerned with the *initial* delays for LSA/SPF
throttling timers, in addition to tuning timers related to link
failure detection. The initial delay for SPF run should be small
enough to allow fast response but long enough to allow LSA flooding
before the SPF run, as OSPF has no special knob (yet) to ensure fast
LSA flooding, like ISIS has. As for other timers, the interval between
consecutive SPF runs should be no less than the time it takes to run
full SPF (from the output of "show ip ospf statistics" or "show isis
spf-log"). You may further space this delay to reduce CPU load for
unstable networks, e.g. if the max-wait=2xSPF_Runtime then your SPF
calculation will not take more than 50% of CPU at any moment. For LSA
throttling, the maximum delay should be no less than the time it takes
network to fully converge (progation delay + spf delay + RIB/FIB
update delay). However, finding these values accurately could be a
problem (e.g. it's hard to find RIB/FIB update delays for all
platforms), so normally all timers are found empirically. Per Cisco's
experimentally validated design:

https://www.cisco.com/en/US/docs/solutions/Enterprise/Campus/routed-ex.html

You may start by setting LSA/SPF throttling timers to 10 100 5000
milliseconds (start/hold/max). This configuration results in subsecond
convergence after a link failure, though the maximum delays could look
a bit high. Additionally, there are much more considerations behind
fast convergene than simply tuning the timers: e.g area sizing,
prefix-suppression, link redundancy and equal-cost multipathing,
multicast RPF throttling, additional OSPF timers tuning (LSA arrival
delay, LSA flooding delay). Lastly, it is important to notice that
IS-IS has slightly more "knobs" for fast performance tuning compared
to OSPF (e.g. LSP fast-flood or priority-driven prefix insertion -
feature only supporeted by OSPF in IOS-XR), but in general you may
tune both protocols to achieve practically the same convergence times.

In addition to the above document, I would recommend you reading the
following research paper:

http://inl.info.ucl.ac.be/publications/achieving-sub-second-igp-convergence-

As for IP Even Dampening timers, I made a short writeup on the timer
selection here http://blog.ine.com/2010/05/03/optimizing-ip-event-dampening/
. The use of IP Even Dampening overlaps with the LSA throttling
feature for ISIS/OSPF protocols, but it could be a very needful
addition to EIGRP networks. It does not hurt using IP Event dampening
with any protocol, as the default 30 second half-life value will
suppress fast oscillation and prevent additional work at
OSPF/ISIS/EIGRP level, but stilll allows fast convergence in response
to the initial event.

HTH,

-- 
Petr Lapukhov, petr_at_INE.com
CCIE #16379 (R&S/Security/SP/Voice)
Internetwork Expert, Inc.
http://www.INE.com
Toll Free: 877-224-8987
Outside US: 775-826-4344
2010/5/22 Hussam EL Kebbi <hussamkibbi_at_hotmail.com>:
> - How can I relate values needed from show ip ospf statistics to values needed
> i throttle:?
> output of show ip ospf statistic(where should I look at exactly)  SPF
> calculation timeDelta T   Intra D-Intra Summ    D-Summ  Ext     D-Ext   Total
> Reason00:10:10   0      0       0       0       0       0       0       R, 00:08:56   0 0       0       0       0       0       0       R, N,
>
> - if I used dampening only with no values, what the default it will take?
> thanks
>
>
>
>> Date: Sat, 22 May 2010 09:52:02 +0000
>> From: joe_astorino_at_comcast.net
>> To: hussamkibbi_at_hotmail.com
>> CC: ccielab_at_groupstudy.com
>> Subject: Re: OSPF Performance
>>
>> Well, you can use "show ip ospf statistics" to figure out how many times
> things have flapped in some amount of time. The statistics you gather from
> that effectively tell you the frequency of your flaps, which you can then use
> to calculate things like the start, increment and max_wait.
>>
>> throttling SPF does ONLY that though -- it delays the running of SPF.
> However, it has a similar IDEA as IP event dampening. The difference is with
> throttling you are delaying the SPF from happening. With dampening you are
> preventing instability by suppressing the advertisements of LSAs when things
> are flapping. Both use the same principal of how long to do that for (using
> exponential decay).
>>
>> So, to answer your second question it depends on what you want to do or what
> you are told to do. Do you want to delay SPF calculations during flaps, or do
> you want to prevent interfaces/networks from being advertised at all during
> flaps, or both?
>>
>> HTH
>>
>> Regards,
>> Joe Astorino, CCIE #24347
>>
>> "He not busy being born is busy dying" -- Dylan
>>
>> ----- Original Message -----
>> From: "Hussam EL Kebbi" <hussamkibbi_at_hotmail.com>
>> To: ccielab_at_groupstudy.com
>> Sent: Saturday, May 22, 2010 5:26:14 AM GMT -05:00 US/Canada Eastern
>> Subject: OSPF Performance
>>
>> Experts,
>> According to ine bog:
> http://blog.ine.com/2009/12/31/tuning-ospf-performance/
>> timers throttle spf start increment max_wait:is used when dynamic adaptation
>> to unstable network topologies.
>> 1 - How can we reflect the unstable network topologies timers (in throttle
>> timers)?for example using: show ip ospf statistics? if so how can we relate
>> output value to timers so we need to put in timers throttle?
>> 2- It says also dampening would suppress events while throttling simply
>> increases the response times:
>> Do we need to use them together?
>> Thanks
>>
>>
>>
>> _________________________________________________________________
>> Hotmail is redefining busy with tools for the New Busy. Get more from your
>> inbox.
>>
> http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:WL
>> :en-US:WM_HMP:042010_2
>>
>>
>> Blogs and organic groups at http://www.ccie.net
>>
>> _______________________________________________________________________
>> Subscription information may be found at:
>> http://www.groupstudy.com/list/CCIELab.html
>>
>>
>> Blogs and organic groups at http://www.ccie.net
>>
>> _______________________________________________________________________
>> Subscription information may be found at:
>> http://www.groupstudy.com/list/CCIELab.html
>>
>>
>>
>>
>>
>>
>>
>
> _________________________________________________________________
> Hotmail is redefining busy with tools for the New Busy. Get more from your
> inbox.
> http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:WL
> :en-US:WM_HMP:042010_2
>
>
> Blogs and organic groups at http://www.ccie.net
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
Blogs and organic groups at http://www.ccie.net