[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ms-ospf
Subject:    Re: determining the newer LSA,and comment on periodic reflooding
From:       Alex Zinin <zinin () AMT ! RU>
Date:       1999-12-03 13:10:51
[Download RAW message or body]


Barry,

I played with it today, injected about 700 ASEs.
That's interesting---the default timer setting
did not affect the refreshment process (well,
at least within 4 refresh cycles), i.e. the difference
between the min and max LSA ages remained
the same (about 10 secs) as just after the first
origination.
But when I set the timer to the minimum possible
value (10 secs) it did work---the first cycle
refresh events were dispersed within about 50
secs. For the second cycle the dispersion was
about 100 secs, and then about 150...

So, seems to be that the documentation is not
quite correct, because according to what is
said in it, the LSA refreshes must be syncing
more and more after each cycle. The real
implementation is different, so I withdraw my
comment.

Padma,
Though I agree that packet pacing is a general
technique, LSA pacing is a link-state technology.
I think the refresh dispersion methods do fit into
the WG, because they improve the protocol
characteristics. At the same time I understand
the situation where a vendor doesn't want to
disclose some details. Since Zebra is GNU GPL,
it's not an issue for us, cisco is a different case :)

Regards,
Alex.



At 09:44 02.12.99 -0800, Barry Friedman wrote:
>>
>> Padma,
>>
>> Yes, I'm aware about these IOS techniques, but
>>
>> 2. From the description I read on CCO, the method
>> used for LSA group pacing seemed to be not perfect,
>> since it doesn't remedy the problem of the whole set
>> of ASEs being refreshed by the ASBR 30 mins after
>> the "redistribute <something>"  was issued (provided
>> that <something> is pretty stable). Am I wrong?
>
>It tries to fix this as well. Each LSA has its own timer
>that has a random jitter value added to its expiry time. The
>idea is that over a long period of time the LSA's start
>refreshing at different times and eventually should
>spread out over the set of group pacing buckets. If the
>database is large you can shorten the group pacing interval,
>which will create a larger number of group pacing buckets. I
>recall Derek telling me that he had run tests to verify that
>this was working.
>
>Barry
>--
>Barry Friedman      mailto:friedman@cisco.com
>Software Engineer/  phone:408-526-8284
>Code Dependent      "Not everything that can be counted counts,
>Cisco Systems Inc.   and not everything that counts can be counted"
>                     -- Einstein
>
>
------------------------------------------------------------------
Alex D. Zinin, Consultant
CCSI #98966
CCIE #4015

[prev in list] [next in list] [prev in thread] [next in thread]