[quagga-users 8035] Re: Quagga / ospfv2 (version 0.99.6) - Reduce convergence time

Preben Myrvoll PMyrvoll at oslo.westerngeco.slb.com
Thu Feb 15 12:23:09 GMT 2007


Hi -

I'm still using Quagga 0.99.6 :-)

* changing to - "timers throttle spf 10 100 5000" - Did not do anything 
(and I've tried various options spf like "100 100 1000").
* Using link-detect does not alter anything (as you pointed out)
* Seting up point-to-point on interfaces in the 11.0.x.x made the ping 
go through after 10 seconds (for both) ( and yes i saw that there was no 
DR/BDR election ;-). With this setup I see that the routers now wait 10 
seconds before sending the second Link State Update (and thus the ping 
in both directions go through after 10+ seconds)

Then ting is - The actual DR - BDR election happens quite fast (with 1 - 
2 seconds). The dead-intervall is still 2 and the hello-interval is 1. 
But the actual LSA exchange / and LSA refresh continues after this, and 
it seems like its the LSA Update/refresh that actually causes the - 
after 5 seconds - DR to be able to ping the BDR, and then 5 seconds 
later the BDR can ping the DR (I'm still talking about pinging the 
"outer addresses" - and with out the point-to-pont)

11.0.11.0/24 <--- eth1| OSPF Router and Machine A| eth0 ---> 11.0.9.0/24 
<--- eth1| OSPF Router and Machine B| eth0 ---> 134.32.79.0/24
- Machine A have ip 11.0.11.1 on eht1 and 11.0.9.250 on eth0
- Machine B has ip 11.0.9.1 on eth0 and 134.32.79.185 on eth0

BTW: What did you mean by:
..
( Idon't see a point in making RouterDeadInterval less than dead time).
..

Thanks again
Preben

Gobbledegeek wrote:
> All
> On second thoughts if the interface goes down, ospf will not try to
> become DR on that segment ... pardon the confusion I created.
>
> I guess the timers throttle is your only hope ...
>
> Regards
> Rahul
>
> On 2/15/07, Gobbledegeek <gobbledegeek at gmail.com> wrote:
>> Hi
>> so  timers throttle spf 200 200 1000 did not work for you? How about
>> "timers throttle spf 10 100 5000" ?
>>
>> There will be three sources of delays here:
>>
>> 1. Time to detect link failure - (do you have linik-detect
>> configured?). Desktop PC's may not provide time granularity below
>> 500ms or one clock tick.( RTOSes on ASICs as in cisco routers will
>> ...)
>>
>> 2.  SPF computation time - timers throttle spf <spf-start> <spf-hold>
>> <spf-max-wait> can reduce this. Other factors are number of neighbors
>> to flood lsa's to, time taken to generate and propogate lsa's
>>
>> 3. Time to update routing tables -  depends on OS will be slow for
>> desktop OSes. (tslking in milliseconds)
>>
>> Did you try setting the RouterDeadInterval to 2 seconds ? Since the
>> Designated router on a broadcast medium will be responsible for
>> generating lsa's, there will be a 2 second delay (default 5 I think)
>> involved here - both will become DRs on a partitioned network ( I
>> don't see a point in making RouterDeadInterval less than dead time).
>> SPF will only run in both routers after a new router lsa is generated
>> with Max-Age (correct me if I am  wrong, I am currently away from my
>> reference library). In this case the major  bottle neck may be here.
>>
>> Alternatively you can try upgrading to 0.99.6 and making the ethernet
>> link point-to-point (0.99.6 has a patch for correctly doing this) on
>> both routers ( interface command) First try with existing version,
>> (command exists but behavior may not be rfc spec - I have used
>> point-to-point mode in 2005 to prevent DR/BDR election). This may help
>> you achieve very short failover times together with "timers throttle
>> spf" command...
>>
>> Regards
>> Rahul Sawarkar
>>
>>
>>
>> On 2/15/07, Preben Myrvoll <PMyrvoll at oslo.westerngeco.slb.com> wrote:
>> > Hi, and thanks for all your assistance:
>> >
>> > In my setup I already have the following params:
>> >
>> > ip ospf hello-interval 1
>> > ip ospf dead-interval 2
>> >
>> >
>> > I've also now tried:
>> >
>> > timers throttle spf 200 200 1000
>> >
>> > Regards
>> >
>> > Any other setusp/configs?
>> > Preben
>> >
>>
>
>



More information about the Quagga-users mailing list