Good day to Pete and everyone on the list.
> Our current plan is to base our upcoming batch of production
> machines on 2.0.36, with future batches base on 2.2.x when it
> reaches "production quality".
> We ran extensive tests on 2.0.36prex and while we have confidence
> with the stability of 2.0.36pre15, we'd very much appreciated if you
> can give us some assurance that 2.0.36 _will_ be an official release
> in the near future.
> We also like to know if there are any known spectacular (and/or
> elusive) 2.0.36 showstopper issue(s) still being left unresolved?
I have been reflecting on this myself. The 2.0.x kernels may not be
ready for production environments depending on your circumstances and
configuration.
I developed the new messaging system in my day job for the local
university. The current deployment state is based on two servers
which are both running Linux. At the stage of 2.0.35 I would have to
answer yes/no with respect to the 2.0.x kernel being ready for a
'production' environment.
The incoming mail/SMTP gateway has performed flawlessly and is
currently sitting with an uptime of around 40 days. It is a simple
PII-233 with 32 meg ram running 2.0.35. It routes anywhere between
175-500 megabytes of mail per day and averages around 30,000 hits/day
to its LDAP server. I haven't heard as much as a peep out of this
machine.
The main IMAP/SMTP server is running a dual PII-300/440FX Natoma with
256 Meg of RAM, DPT 1334 SCSI/RAID 5 and is currently running with a
uni-processor 2.0.35 kernel updated with Donald's 1.04 driver for the
SMC Etherpower II.
This particular machine averages around 120-130 simultaneous IMAP
connections during the day. At peaks we see 200+ connections. Along
with handling sendmail runs I would judge this server to be under
moderate load. We see loads of 6-10 during sendmail deliveries.
In this particular environment/configuration the 2.0.35/SMP kernel is
not stable. Performance has been good but stability could not be
considered production quality yet.
In around 35 days of operations we have seen 3 system deadlocks
secondary to the boot CPU interrupt deadlock problems that Leonard
Zubkoff documents in kernel/sched.c. More problematic have been the
networking hangs caused by the SMC Etherpower II's becoming 'stuck'.
This has been remedied in all cases by upping and downing both
interfaces. We were up to about 2-3 of these situations a week when I
backed away from the SMP kernels.
With uni-2.0.35 on this machine we experienced one situation of
'stuck' network interfaces. I realized that the v0.99 driver module
was loaded so I compiled v1.04 from Donald's site and we haven't seen
any problems since than although it has only been about 24 hours.
I tried 2.1.125 SMP and we managed to last about 4-6 hours until load
got heavy. The network interfaces hung and although upping and
downing them restored functionality the primary interface refused to
function properly. There were wildly variant ping times resulting in
unacceptable network performance. Incidentally this was 2.1.125 with
the v1.04 drivers.
I think that 2.0.35/SMP would be suitably stable for a production
environment except for issues with the networking drivers that we are
using. I am planning to look at the work that Linus has done on the
eepro drivers in 2.1.126 to see if I can glean any SMP locking
information that may be helpful in stabilizing the epic drivers. I am
suspicious of SMP issues since the un-PII-233 mail gateway which has
been trouble-free is using an SMC card out of the same carton as the
SMP machine.
I guess the point of this note is that stating that 2.0.x is a
production kernel has to be delivered with a sense of caveat. In our
shop a machine being production level implies SMP. Getting a stable
SMP configuration in 2.0.x is still somewhat dependent on environment
and configuration, especially with respect to drivers.
> Respectfully,
> Pete
> teamwork@freemail.c3.hu
Have a pleasant weekend everyone.
Greg
}-- End of excerpt from teamwork@freemail.c3.hu
As always,
Dr. G.W. Wettstein Enjellic Systems Development - Specialists in
4206 N. 19th Ave. intranet based enterprise information solutions.
Fargo, ND 58102 WWW: http://www.enjellic.com
Phone: 701-281-1686 EMAIL: greg@wind.enjellic.com
------------------------------------------------------------------------------
"Linux is a movement, a philosophy, where programmers and technical
people take control of their own destiny."
-- Tim Bass
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/