[c-nsp] 7500 having problems after upgrade from 12.0.19S2 to 12.0.31S1

[Robert E.Seastrom]

Rodney Dunn <rodunn at cisco.com> writes:

> On Tue, Nov 22, 2005 at 08:14:42AM -0500, Robert E.Seastrom wrote:
>
>> ST merge was after 12.0(21)S.  The new BGP data structure is said to
>> be more computationally efficient at the expense of memory efficiency.
>
> Absolutely correct. However, there is a memory gain too in that
> for a full reconvergence it takes much much less transient memory
> to reconverge.

Wasn't aware of that.  Interesting.

> Not pointed at Robert but a general comment:
>
> We can't code new features without making the image larger.
> We do (even though some may not think so) try very very hard to
> be extremely memory conscious when we code new features.

I'm with you here.  By definition, adding features takes memory, and
if my understanding is correct, there's already enough shared code
between certain features that ripping out something we don't care
about anymore (for instance, X.25) would not gain much unless we took
out other stuff (for instance, FR) that we still want.

I think all in all Cisco has done a pretty good job of avoiding code
bloat (reasonable people will disagree as to whether the same
assessment applies to "feature list bloat" as well.  That's a
discussion for another day, though I'm sure I'm not alone in wishing
fervently that certain old releases would continue to be supported).

> The
> mistake we've made is having parts where they memory can't be
> upgraded and we are trying to not make that mistake again.

If this is well-understood inside of Cisco, particularly in the
hardware design arena, then I'm pretty happy going forward.

> 256M is simply not enough anymore for dual feeds and recent code.

Even with old code, the end is in sight.  Dual feeds on a GSR with
post-merge 12.0S and not much in the way of an IGP only has 31 megs
free.

                                        ---Rob