Re: regarding ospf las flushing .....

[Dave Katz <[e-mail]>]

The tuple of originating router, LSA ID, and LSA type *always* uniquely 
and completely identifies an LSA, by definition.

There can never be any ambiguity about which LSA is being flushed, even 
when only the header is sent.  The only way this could even 
conceptually be an issue is if an implementation thought it could send 
two different external routes with the same LSA ID but different masks, 
but this is not allowed (I seem to recall that there was evidence that 
some misguided implementation attempted such a thing recently.)

I wrote the Juniper implementation this way back in '97 and things 
worked swimmingly until a broken implementation came along that tried 
to access the contents of the LSA before noticing that it was being 
flushed, and did something bad (like crashed.)  I don't recall offhand 
how this was resolved, though I'm sure I resisted changing our 
implementation.

--Dave

On Jun 8, 2005, at 4:58 PM, Don Goodspeed wrote:

> Padma,
>
> How does one address the issues of summary and external
> LSAs where the LSID "may" change depending upon Appendix
> E of RFC2328.  In those cases, isn't including the subnet
> mask needed?  Otherwise, an incorrect LSA could be flushed.
>
> -don
>
> -----Original Message-----
> From: [e-mail]
> [mailto:[e-mail]] On Behalf Of Padma 
> Pillay-Esnault
> Sent: Wednesday, June 08, 2005 4:52 PM
> To: [e-mail]
> Cc: [e-mail]; Padma Pillay-Esnault
> Subject: Re: regarding ospf las flushing .....
>
> Anup
>
> This was already done in a major implementation. I think it was a good 
> idea.
>
> You have to be careful though as this might break some implementation 
> who
> access the body of the lsa on flushing ( though I don't see why they
> would do that).
>
> I'm for it and can collaborate. This can initiate a discussion on the 
> list.
>
> Padma
>
> anup wrote:
>
>> Hello Padma,
>>
>> As per RFC 2328, we send the lsa (header + body) to the peer though
>> the lsa is maxaged.
>>
>> Considering that the peer would not examine the lsa body if the lsa is
>> maxaged, *if we could send only the maxaged lsa's header*, it would
>> reduce a lot of traffic as well as the protocol memory consumption
>> during flushing.
>>
>> If you agree with this idea, I would like to prepare a small draft on
>> this.
>>
>> Regards,
>>
>> Anup
>>
>