Re: Hitless Restart 02 : 2-way, BDR, grace policy

Erblichs <erblichs@EARTHLINK.NET> Tue, 21 May 2002 15:09 UTC

Received: from cherry.ease.lsoft.com (cherry.ease.lsoft.com [209.119.0.109]) by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA13877 for <ospf-archive@LISTS.IETF.ORG>; Tue, 21 May 2002 11:09:38 -0400 (EDT)
Received: from walnut (209.119.0.61) by cherry.ease.lsoft.com (LSMTP for Digital Unix v1.1b) with SMTP id <2.00614F3B@cherry.ease.lsoft.com>; Tue, 21 May 2002 11:09:55 -0400
Received: from DISCUSS.MICROSOFT.COM by DISCUSS.MICROSOFT.COM (LISTSERV-TCP/IP release 1.8d) with spool id 877981 for OSPF@DISCUSS.MICROSOFT.COM; Tue, 21 May 2002 11:09:55 -0400
Received: from 207.217.120.22 by WALNUT.EASE.LSOFT.COM (SMTPL release 1.0d) with TCP; Tue, 21 May 2002 11:09:54 -0400
Received: from 209-239-195-16.oak.jps.net ([209.239.195.16] helo=earthlink.net) by hawk.mail.pas.earthlink.net with esmtp (Exim 3.33 #2) id 17ABGq-0001kf-00 for ospf@discuss.microsoft.com; Tue, 21 May 2002 08:09:52 -0700
X-Sender: "Erblichs" <@smtp.earthlink.net> (Unverified)
X-Mailer: Mozilla 4.72 [en]C-gatewaynet (Win98; I)
X-Accept-Language: en
MIME-Version: 1.0
References: <3CDDA283.9661130D@earthlink.net>
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Message-ID: <3CEA6554.5631C7CE@earthlink.net>
Date: Tue, 21 May 2002 08:18:44 -0700
Reply-To: Mailing List <OSPF@DISCUSS.MICROSOFT.COM>
Sender: Mailing List <OSPF@DISCUSS.MICROSOFT.COM>
From: Erblichs <erblichs@EARTHLINK.NET>
Subject: Re: Hitless Restart 02 : 2-way, BDR, grace policy
To: OSPF@DISCUSS.MICROSOFT.COM
Content-Transfer-Encoding: 7bit

Group,

        I got a couple of private replies and wanted
        to clarify a few items.

        IF the system memory could be checkpointed/saved
        AND later resumed upon restart,

        then the main issues become,

           * what is the minimum of the OSPF interfaces
        DeadInterval? This is the value that the system must
        be re-booted so none its neighbors don't declare it as
        dead and free its associated resources.

            * can you save the memory in a format that
        upon resumption you could restore it in newer
        data structures? i.e. a new release.

           * how do you save the hw/sw registers, that upon
        bootup, the system behaves properly?

        Etc..

        Then grace-lsa's would not be necessary, no knowledge
        of the neigbors that you are about to perform this
        procedure would be necessary, and it would be
        backward compatible with neighboring routers.

        This work was done at Terayon a couple of years
        ago and I called it "Fast Re-start". This general
        feature was also done at Sun Microsystems and they
        call it checkpoint/resume.

        Mitchell Erblich
        ======================
        PS:  #8 from below would partially be satisfied if
        MaxAge flushing was explicitly specified for grace-LSAs..

Erblichs wrote:
>
> Hitless Restart -02 by J. Moy : Feb02 to Aug02
>
>         Pre-note: I have NOT previously done this level of
>         restart with a grace-LSA and am not currently working
>         on this type of functionality..
>
>         My assumptions are that I understand this draft and
>         their are a few things that may be a bit clearer...
>         Sorry ahead of time...
>
>         1) Overview 1.:
>            Should their be an explicit statement that 2-way
>            DRothers link partners, not not play a role in
>            hitless restart?
>
>         2) Overview 1.:
>            If you are a DRother, what does this really gain
>            you?
>
>         3) Is restart time started when the first grace-LSA
>            is sent out OR when the last grace-LSA is acknowledged?
>
>         4) 2) Op of restart router (3) If the restarting ...
>            Couldn't you use nvram to store that you were the
>            DR? You then wouldn't have to wait 1/2 on average
>            of the hello interval.. Complication, what happens
>            with demand circuits and suppressed hellos?
>
>         5) Same section as #4.
>            How do you preserve the fact that you were the BDR
>            in this Draft? I actually don't really see any mention
>            of the BDR...
>                 -- My suggestion in #4 would work as well for a
>                    BDR or a DRother..
>
>         #4 and #5) For consistency sake, these items should then
>           be checked with hellos, etc..
>
>         6) 2.1) (2) What happens if the nvram cypto seq # somehow
>         is no-longer correct after the restart? I assume this
>         can be changed.
>
>         7)  2.1) A grace-LSA is originated ..
>            Am I correct in assuming that the below question would be
>         implicitly covered as a topolgy change and proceed to normal
>         restart?
>
>         What happens if a BDR comes up during restart (before all
>         grace-LSA are acknowledged?
>
>         8) 2.1) A grace-LSA is originated ..
>         A) Conflict with 3.1 (4)b. How do you resolve a grace-LSAs that
>         were xmit'ed and acked by one or more neighbors and one
>         or more grace-LSAs that are in conflict with exceeding the
>         local policy of time T.
>         B) How do you detect that you are violating time T?
>         C) Can you resend a grace-LSA with a different grace-period?
>            What should a router do, if two grace-LSAs were sent
>            with the same grace-periods? With different grace periods?
>         D) What happens if you violate your grace-LSA value?
>         E) What happens if the grace-period is exceeding long and
>            router X was the DR?
>         F) When do you stop re-xmiting grace-LSAs and proceed to
>            normal restart? Can you spend you entire grace-period
>            doing re-xmits?
>
>         9) 2.2 (2) add a d) router that would be a helper candidate
>         came up after restart was started and all grace-LSAs were
>         acknowledged.
>
>         10) Why not save xmit OSPF hellos per interface in nvram,
>          so after a restart prior restart hellos can be reloaded
>          and sent?
>
>         11) #4) "or they do not recieved the grace-LSA"
>                 I don't understand this phrase..
>             Are you talking about not reciving a ack from a router
>             Y candidate?
>
>         Thanks,
>                 Mitchell Erblich