[j-nsp] Need suggestions..

[Richard A Steenbergen]

On Thu, Feb 04, 2010 at 12:41:31PM +0200, Pekka Savola wrote:
> Well, I don't know about i's, but FWIW plain old M10 runs e.g. RE-600 
> (2G memory etc.) just fine.

I still have an M10 floating around (doing a chds3 termination for a
handful of t1's, nothing special), and rest assured it can NOT reliably
take a full table when running even vaguely modern code (9.4), even
after we upgraded the DRAM on the FEB. After a few weeks it would start
dropping routes it was trying to install to the PFE, occasionally
blackholing traffic in the process, and requiring a reboot to fix. The
only solution was to stop taking a full table.

The only way this could possibly work is if you can get away with
running some really old and unsupported code, from back in the day when
JUNOS wasn't a giant buggy bloated pile of crap. Of course, they also
aren't maintaining security fixes for the really old versions, so the
next time something like the last batch of issues comes along you may be
left out in the cold. I also have a huge pile of fwdd.core files from a
J2300 running 9.3R4 (the last release of code for old J-series hw) from
my attempts to use it as an IPSEC endpoint over the last couple weeks,
which prove Juniper isn't even leaving you with a reliable/working
"final image" before they pull the plug. IMHO it would be damn near
suicidal to think you can get away with running an old EOL platform like
the M10 in production and expect future code to work on it reliably,
regardless of RE performance (and the p3 600MHz isn't really that much
faster anyways :P).

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)