How many users are in a given machine?

[Djin]

Like, for example, Hydrogen?

[JPC-Les]

Perhaps its best to focus on server load rather than the number of occupants. Right now this is the load of hydrogen:

11:08am up 7 days, 17:30, 1 user, load average: 0.33, 0.34, 0.31

Our goal is to keep the server load at or around 2.0 (dual CPU remember).

[edit]

Actually to better answer your question we really dont focus on the number of clients per server. Our main focus is on how each client affects the server. One client can tax the server beyond its capabilites while at the same time you could have up to 1000+ clients on a single machine happily if they only used minimal system resources.

[Vin DSL]

Originally posted by Les
...its best to focus on server load rather than the number of occupants... Our main focus is on how each client affects the server. One client can tax the server beyond its capabilites...

This is the correct way to do things, however, it begs the question --- do you have any way of determining this?

[Gwaihir]

Les, can you tell us what those numbers 0.33, 0.34, 0.31 mean?

As well as what you mean by 2.0? First thought I get from your decription for 2.0 is both CPU's fully loaded, but I do hope that's not what you mean..

[JPC-Les]

Originally posted by Vin DSL
This is the correct way to do things, however, it begs the question --- do you have any way of determining this?

Yes, there are tools and methods to do this.

Originally posted by Gwaihir
Les, can you tell us what those numbers 0.33, 0.34, 0.31 mean?

As well as what you mean by 2.0? First thought I get from your description for 2.0 is both CPU's fully loaded, but I do hope that's not what you mean..

The numbers are current CPU load, 5 minute CPU load average, and 15 minute CPU load average.

Your impression is correct and generally there is no problem having a load near 2.0. The server will preform correctly and its not until the load reaches a load above its capabilities that problems arise.

[Vin DSL]

Originally posted by Les
Yes, there are tools and methods to do this...

Next question: Site 'X' is discovered to be gobbling up system resources for an extended period of time. What is the fix?

[JPC-Les]

Originally posted by Vin DSL
Next question: Site 'X' is discovered to be gobbling up system resources for an extended period of time. What is the fix?

That depends on how and why said system resources are being "gobbled up". Theres quite a few ways to tax a server. The fix can be a simple tweak to a script or asking the client to upgrade to a semi-dedicated or dedicated solution.

[Darrin]

to talk about sever load just a bit, a server loat of 1.0 means the server is performing as "loaded" as it can without noticing any slowing. Since they have dual CPUs you DOUBLE that number, so therfore the MAXIMIM "load" lydrogen can have without noticing performance idfferences is 2.0, if you are running in the .3's you are doing GREAT!!

[Djin]

How can I find the server load or the number of clients manually?

I'd like to see who's causing the most load on my server...

[Coyote]

Dont know about fining out number of clients - as we are Jailed now, but server load is easy.

Just log in via SSH and typethe command: uptime

[Gwaihir]

But I'd expect huge ups and downs in it, right? So that "2.0" isn't bad for the first mark (this minute) but would be bad as a 15 minute average?

For example, I got these numbers today (Oxygen 15:43 UTC):

load average 4.47, 2.44, 1.38

which I'd interpret as 'filled to capacity', maybe even a bit over the top; certainly not below.

Five minutes later it went up to 'auch':

load average 7.40, 3.71, 2.07

and another five minutes later (the shell is now noticably slow to respond):

load average 7.01, 5.92, 3.54

I think you guys need to have a look at this..

[edit]Should I put in a support ticket when I notice this happening?[/edit]

[jason]

Numbers that high look like a runaway proces--probably a user's script caught in an endless loop or something. I've never seen numbers that high on any properly running server (JPC or otherwise).

Occasionally server loads can spike if a lot of users try to connect at the same time or whatever, but to consistantly see numbers up high, especially for the 15 minute average, seems to indicate a problem.

--Jason

[Gwaihir]

I figured as much. They went on further up and up, so I put in a ticket. They've upped the montoring on the server, but didn't find the problem.

I guess the response time for tickets will rarely be fast enough to figure one of these out, unless maybe a log catches it. Maybe phone support would help there?

Well, at least it's back to (more) normal now.

[Vin DSL]

Originally posted by Gwaihir
...They've upped the montoring on the server, but didn't find the problem..

Did somebody get moved there from the Pee-Pee server?