Re: Linux can't stay up for more than an hour?

Matthias Sattler (m_sattle@informatik.uni-kl.de)
Fri, 7 May 1999 00:46:44 +0200


Hiho Ben...

On Wed, May 05, 1999 at 05:00:52PM -0400, Ben 'The Con Man' Kahn wrote:
>
> I'm hammering a Linux machine by placing a high volume
> (1,000,000 hits/day, 500,000 CGI hits/day) web site on it. The box can

This would mean 11.5 hits/second and 5.8 CGI hits/second if the load is
equally distibuted along the day. But usually it isn't. So you easily get
double the numbers or even more hits during the business hours.
20 or more hits/second shouldn't be a problem for your machine and your
network should do that too, but 10-20CGI bins per second may be a problem.
What kind of CGI bins do you have? perl? I've measured the startuptime of
perl on my (slow) machine when it is already in the diskcache. It says 0.03 sec
(/usr/bin/time /usr/bin/perl < /dev/null). Let's say that your machine can do
it in 0.01 sec. So 20 perl startups already take 20% of the time of one CPU.
Now keep in mind that the perl scripts are interpreted and way slower than
compiled code. It is possible that even with 2 CPUs your computer is
underpowered. Think about buying xeons instead of normal PIIs. But I would
recommend that you rewrite your CGIs in C and have a look what happens then.

> Memory: 128MB

128MB may be not enough.

> Disks: 9Gig local, 12Gig remote NFS from Irix 6.5.2 and locking on.

Do you serve remote files? If so, this may be a problem.

> knfsd: 981204-1

Have you tried the normal nfsd? I think it can handle file locks too.
By the way, knfsd is the nfs server not the client, do you export
things too? For the nfs mount from Irix the nfs client is important.
What mount parameters do you use?

>
> The machine dies so hard that the screen goes black and does
> NOTHING. Nothing on the screen, no keyboard input is accepted, no network
> access is accepted. When we recycle the machine, it usually won't restart

Do the keyboard leds still work then (in textmode not under X)?

> because of file corruption and we need to get someone to the box and run
> fsck on it.
>
> We tried just recently to put this on a different Linux box. We
> have the exact same software configuration on this other box, but it's a
> single processor AMD-K6 333. This machine can't handle the load, and
> eventually dies in the same way. (As it dies, loads are 80+.)
>
> Please help! We're going to switch to BSD even though NFS locking
> doesn't seem to work...

If your machine is underpowered, switching to another OS won't help. If it's
a nfs problem (like you suspect in one of the following mails) it might.
Hmmm... let's start debugging.

Matthias

---

O .---------------. .___________. O /\/ . `. m_sattle@ ,' / \ +FAX . \/\ __..--- ' /\/ | `._________,' | (___)/ * * \(___) \/ \ ---..__ ""---__ \/`. | informatik. | / | \ +49 (0)6333 ,'\/ __---"" .. / | .uni-kl.de | | `...' | -65079 \ ...' `---------------' `._____.'

--> Don't take life too seriously -- you'll never get out of it alive. <--

- 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/