JPL config file

[Randy Terbush <ra...@zyzzyva.com>]

I've exchanged a few emails with one of the responsible people at 
JPL. He indicates the following:

> Our web site is split across two servers, www-a.jpl.nasa.gov and
> www-b.jpl.nasa.gov.  These are both mirrors of a third machine,
> www-onlab.jpl.nasa.gov.  The www-onlab machine is restricted to
> JPL internal access only.  Www-a and www-b share the load of
> accesses to www.jpl.nasa.gov via round-robin DNS.  All of our
> machines are Sun Ultra 1's (170 MHz) with 64 MB of memory and
> something like 11 gigs of disk each.  We're running Apache httpd
> 1.2b7.

The first problem looks like a shortage of memory, but he goes on 
to say that the CPU is 90% idle. I have asked about listen queue 
size but don't have that data yet. Being Solaris, they might 
benefit from 1.2.1.

Relevant config is below. Basically a stock httpd.conf file with 
comments removed (by me). They could probably safely raise the 
MaxRequestPerChild value, but based on the CPU load, I don't think 
that it is forking a lot.

I've not made any recommendation other than memory at this point. 
If someone here would like to contact them directly, send me 
private email.

------- Forwarded Message

ServerType standalone

Port 80

HostnameLookups off

ScoreBoardFile logs/apache_runtime_status

ServerName www.jpl.nasa.gov

Timeout 1200

MinSpareServers 100
MaxSpareServers 200
StartServers 100 
MaxClients 512
MaxRequestsPerChild 50

------- End of Forwarded Message

Re: JPL config file

[Rob Hartill <ro...@imdb.com>]

On Sun, 6 Jul 1997, Randy Terbush wrote:

> ------- Forwarded Message
> 
> ServerType standalone
> 
> Port 80
> 
> HostnameLookups off
> 
> ScoreBoardFile logs/apache_runtime_status
> 
> ServerName www.jpl.nasa.gov
> 
> Timeout 1200

If they are stuck for processes to answer requests, booting a few of the
slow clients off is a good thing. 20 minutes ? anyone would think the
link is directly to Mars itself. 2 minutes is probably more than enough.

> MinSpareServers 100
> MaxSpareServers 200
> StartServers 100 
> MaxClients 512
> MaxRequestsPerChild 50
> 
> ------- End of Forwarded Message


does this make us rocket scientists ?


--
Rob Hartill                              Internet Movie Database (Ltd)
http://www.moviedatabase.com/   .. a site for sore eyes.

Re: JPL config file

[Marc Slemko <ma...@worldgate.com>]

Yup, his biggest problem is memory.  He may want to bump down MaxClients a
bit; he could get better performance if the machine isn't swapping to
hell.  

Compiling with the fewest possible modules could help the memory a bit.

Unfortunately, if I were him with such resources (64 meg is nothing,
especially on a RISC box) my best solution would be Zeus.  Apache performs
well, but one process per request just can't compare to one process for
everything if you don't have the memory.

I note he has a ScoreBoardFile directive... verify that it is compiled to
MMAP it (in which case ScoreBoardFile does nothing), as is the default for
Solaris, if that is what is running. 

On Sun, 6 Jul 1997, Randy Terbush wrote:

> 
> I've exchanged a few emails with one of the responsible people at 
> JPL. He indicates the following:
> 
> > Our web site is split across two servers, www-a.jpl.nasa.gov and
> > www-b.jpl.nasa.gov.  These are both mirrors of a third machine,
> > www-onlab.jpl.nasa.gov.  The www-onlab machine is restricted to
> > JPL internal access only.  Www-a and www-b share the load of
> > accesses to www.jpl.nasa.gov via round-robin DNS.  All of our
> > machines are Sun Ultra 1's (170 MHz) with 64 MB of memory and
> > something like 11 gigs of disk each.  We're running Apache httpd
> > 1.2b7.
> 
> The first problem looks like a shortage of memory, but he goes on 
> to say that the CPU is 90% idle. I have asked about listen queue 
> size but don't have that data yet. Being Solaris, they might 
> benefit from 1.2.1.
> 
> Relevant config is below. Basically a stock httpd.conf file with 
> comments removed (by me). They could probably safely raise the 
> MaxRequestPerChild value, but based on the CPU load, I don't think 
> that it is forking a lot.
> 
> I've not made any recommendation other than memory at this point. 
> If someone here would like to contact them directly, send me 
> private email.
> 
> ------- Forwarded Message
> 
> ServerType standalone
> 
> Port 80
> 
> HostnameLookups off
> 
> ScoreBoardFile logs/apache_runtime_status
> 
> ServerName www.jpl.nasa.gov
> 
> Timeout 1200
> 
> MinSpareServers 100
> MaxSpareServers 200
> StartServers 100 
> MaxClients 512
> MaxRequestsPerChild 50
> 
> ------- End of Forwarded Message
> 
> 
> 

Re: JPL config file

[Dean Gaudet <dg...@arctic.org>]

You asked them to upgrade to 1.2 right?  The networking and other
performance improvements went in after 1.2b7.

Dean

On Sun, 6 Jul 1997, Randy Terbush wrote:

> 
> I've exchanged a few emails with one of the responsible people at 
> JPL. He indicates the following:
> 
> > Our web site is split across two servers, www-a.jpl.nasa.gov and
> > www-b.jpl.nasa.gov.  These are both mirrors of a third machine,
> > www-onlab.jpl.nasa.gov.  The www-onlab machine is restricted to
> > JPL internal access only.  Www-a and www-b share the load of
> > accesses to www.jpl.nasa.gov via round-robin DNS.  All of our
> > machines are Sun Ultra 1's (170 MHz) with 64 MB of memory and
> > something like 11 gigs of disk each.  We're running Apache httpd
> > 1.2b7.
> 
> The first problem looks like a shortage of memory, but he goes on 
> to say that the CPU is 90% idle. I have asked about listen queue 
> size but don't have that data yet. Being Solaris, they might 
> benefit from 1.2.1.
> 
> Relevant config is below. Basically a stock httpd.conf file with 
> comments removed (by me). They could probably safely raise the 
> MaxRequestPerChild value, but based on the CPU load, I don't think 
> that it is forking a lot.
> 
> I've not made any recommendation other than memory at this point. 
> If someone here would like to contact them directly, send me 
> private email.
> 
> ------- Forwarded Message
> 
> ServerType standalone
> 
> Port 80
> 
> HostnameLookups off
> 
> ScoreBoardFile logs/apache_runtime_status
> 
> ServerName www.jpl.nasa.gov
> 
> Timeout 1200
> 
> MinSpareServers 100
> MaxSpareServers 200
> StartServers 100 
> MaxClients 512
> MaxRequestsPerChild 50
> 
> ------- End of Forwarded Message
> 
> 
> 
>