Still Considering Debian - But Stuck!
Hi Everyone,
A while back I asked for some feedback and got a very rich set of info
from folks about Debian used in a stable ISP environment as compared to
other OS's and distributions. All the info was very helpful and helped
us further solidify our desire (though not yet decision) to make Debian
our platform as we move forward.
We've run into a couple rather HUGE issues, though, that I'd like to get
further feedback on. Not that I couldn't figure it all out for myself,
but nothing beats someone else's experience when it comes to saving me
the time and heartache ;-) Just about everyone warned me that the
stable Debian distribution would be old and well tested/maintained, but
I'm not sure I was prepared for just HOW old...
Our company uses Java --- a LOT of Java. We therefore use a lot of
threads, and a lot of threads. And a whole mess of threads, too. Under
Red Hat 7.3, we found that when the system had a total of say, 10,000
PID's given out (nearly all of them to threads) the system would become
very unstable. When we moved to Red Hat 9 for the affected systems,
which includes the new 0(1) scheduler, and either a different kind of
thread support in either the kernel or GlibC, this problem went away.
I'm honestly not sure who is responsible for the way threads are
handled, and I suspect it's not exclusively the kernel, but under RH9
each JVM (or any app with threads) gets a single PID as normal and all
very strange behavior that we saw under RH7.3 disappears.
I see that Debian 3.0r2 includes a nicely aged (like fine cheese) Linux
2.2 kernel. While I'm certain the aging process only makes its flavour
stronger and more delectable, I'm afraid it's going to choke at the
thought of 10,000 threads. Say nothing of 20,000. Now I imagine it's
not so difficult to simply compile a recent 2.4 (2.5?) kernel and go
from there. Is this fair? Or would you suppose that the current stable
Debian is too old in other areas to properly handle kernel 2.4?
Even if I replace the kernel, I'm concerned that there's more involved
with the more efficient handling of threads from RH 7.3 to RH 9 than
just a kernel change -- I have to think there was a significant rework
of some libraries that made threads more efficient under RH9 as well.
Would anyone be able to identify exactly what that re-working was, and
conjecture if they think it can be done under 3.0r2? For that matter,
would I at that point be running so much new technology that I may as
well be running an unstable distribution of Debian?
Finally, while I'm messing around with the kernel, I'd have to include
support for ext3fs. In our environment, journaling is not an option,
it's a base requirement. Of course replacing the kernel would pretty
much give me kernel-level support for it. From that point, how
complicated is it to get the rest of the tools to play nicely with
ext3fs? I'd imagine that a large set of tools would need to be
replaced, including e2fsck, mount, umount, etc.
Thanks once again for all the info so far!
-Fred Whipple
Reply to: