For those webhosting company owners out there, what's the average server load for one server and what's the performance at that level? (machine sluggish or running fast and smooth?)

I'm new with hosting machines (in fact, no experience at all :D ) and was just wondering. I heard that HP-UX servers gets tested for up to 500 currently which seems really high. :eek: Also, what kind of machines do you use? I saw Celerons and P4's at Rackshack but isn't that for Windows? Why would you use Windows machines for Unix hosting? (I'm guessing Linux is installed... but wouldn't you rather use a Unix machine for Unix hosting?)

Wow, look at all those question marks. I'm just making an attempt to learn more about webhosting here.... :blush:

:confused: What is a Linux machine?

Some companies do use Sun UltraSparc servers.

And, I think a couple use SGI Webforce. Atleast, Webcom used SGI back when I was customer.

haha, what is a windows machine? :) something with a WIN-MODEM? :)

I'm also interested in hearing peoples server loads.

Ours usually run between 1 and 2 but that can double in a second, the server (Redhat Linux 7.2, Plesk) works fine with this load.

However we recently had a user install some poorly written scripts that pushed the load up to 160! This made the server totally unusable for everyone. Luckily we caught this quickly ;)

I'm getting the vibe there's no such thing as a Windows machine. *insert sheepish grin here*

Server loads of 1 or 2 is average?? :eek3: How come the HP-UX's can handle loads of 500?

BTW, I heard SGI went kaput....???

500 what? :)

The servers at rackshack come with red hat linux.
They are just machines with operating systems on it. Instead of claling it a windows machine or linux machine, you can call them "compaq machines" "dell machines" "white boxes - generic" etc.

1 or 2 is ok.

One of our servers go at between .5 and 2. Mostly around the .9 and 1.1 area. All our other servers are undred .5 load

Our first server (that had between .5 and 2, used to have loads of 3 and 4 when we had some CPU intensive scripts running. I didn't notice a decrease in performance. I did notice perforemance decreases when load went over 10. This was on an Duron White Box 1ghz.

I watch our servers like a hawk to make sure that they never become overloaded. Our servers average around 0.25 - 0.40 during most peak times.

I prefer to keep them under 0.25, but since we have a lot of Java Tomcat clients, that's hard to do. This server (web2.assortedinternet.com) has over 300 concurrent processes running 24/7 but the load is almost always under 0.30.

bash-2.04$ ps -Aef | wc -l
320
bash-2.04$ w
10:38pm up 40 days, 1:03, 2 users, load average: 0.14, 0.27, 0.27
USER TTY FROM LOGIN@ IDLE JCPU PCPU WHAT
root pts/0 pool-138-88-133- Tue 9pm 0.00s 0.42s 0.16s w
root pts/2 pool-138-88-133- Tue 9pm 1:13m 10:11 0.22s -

It has about 60 accounts on it, 10 of which are Java enabled.

I always investigate when load is over 1 for any period of time for a server. If load reaches more than 5, I start recycling processes to calm things down. Almost every time we've seen a server over 5 it was due to a) a spammer, or b) a runaway CGI Script.

Take care,

Brian

Originally posted by bteeter
I prefer to keep them under 0.25, but since we have a lot of Java Tomcat clients, that's hard to do. This server (web2.assortedinternet.com) has over 300 concurrent processes running 24/7 but the load is almost always under 0.30.

bash-2.04$ ps -Aef | wc -l
320

How many of those 320 are actually just threads from the JVM(s)? I know under linux, Tomcat spawns threads that look like processes but really aren't. Under FreeBSD, it just does one process. It has something to do with the way Tomcat is executed. I can force Tomcat to spawn multiple threads (that look like processes) under FreeBSD, but then things don't work quite right, whereas under Linux it works just fine.

I don't feel the # of processes affects the load at all. It only depends on how many of those processes are _actually_ running (or waiting to run). You can spawn 300 httpd processes and have a load of 0.00. :) They're all just sitting there waiting for connections.

To stay on topic: the load simply means how many processes are waiting to RUN. If you have a load of 500, that means there are 500 processes wanting to RUN. Sure, you can have 500 processes that print "hello world" 5000 times to 500 different files all run at one time and the machine will be likely be ok. Try having 500 httpd processes trying to serve data all at the exact same time for a long period of time. That 500 load will hurt the machine. Try having 500 perl processes trying to process stats data on 500 different large web access logs all at the same time. See what happens then. :)

I think a consistent load of 2.00-4.00 is ok. Anything higher than that is not healthy. Just get a new server to put future clients on.

It's late, I hope all that made sense. :)

Here are some load averages from diffrent machines:

11:26pm up 41 days, 19:58, 1 user, load average: 0.31, 0.14, 0.09
11:28pm up 1 day, 4:59, 1 user, load average: 0.07, 0.12, 0.09
9:31pm up 21 days, 13:57, 1 user, load average: 0.85, 0.92, 0.95
11:30pm up 2 days, 22:11, 2 users, load average: 2.08, 1.95, 2.09

Originally posted by AcuNett
The servers at rackshack come with red hat linux.
They are just machines with operating systems on it. Instead of claling it a windows machine or linux machine, you can call them "compaq machines" "dell machines" "white boxes - generic" etc.

1 or 2 is ok.

One of our servers go at between .5 and 2. Mostly around the .9 and 1.1 area. All our other servers are undred .5 load

Our first server (that had between .5 and 2, used to have loads of 3 and 4 when we had some CPU intensive scripts running. I didn't notice a decrease in performance. I did notice perforemance decreases when load went over 10. This was on an Duron White Box 1ghz.

You also have to consider if the box is dual cpu or not. Top does not incorporate that into its load report so you need to effectively divide the load by two on a dual cpu machine. Which brings up a question, why doesn't someone code top to incoporate dual cpu configs?

Originally posted by Jag


You also have to consider if the box is dual cpu or not.
Yes, this is a good point, CPanel can report high loads if it hits 2, however a Dual CPU can handle well above that without any problems.

Originally posted by rochen

Yes, this is a good point, CPanel can report high loads if it hits 2, however a Dual CPU can handle well above that without any problems.

That brings up another question. When cpanel shows a server load of 1 or above the light goes red.

It does this nightly when it gzips all the files and backs up to the second drive. Once it finshes the load drops back to .3 or so. What controls the threshold of when that light turns red?

It scares some of my customers, although there is no noticable slowdown in the system at all.

Originally posted by Monte


That brings up another question. When cpanel shows a server load of 1 or above the light goes red.

It does this nightly when it gzips all the files and backs up to the second drive. Once it finshes the load drops back to .3 or so. What controls the threshold of when that light turns red?

It scares some of my customers, although there is no noticable slowdown in the system at all.

The cpanel status code is designed to show that red light as warning but in my opinion the bar should be raised to 5 before it goes red. We use all dual cpu systems though but half the time those status lights get stuck on a certain color forever even after the load is down or a server that was down comes back up.

Funny that you guys mention that. I was just thinking about 20 minutes ago about how silly it is to have that blinking red button at a load above 1. Especially when it's accessible from the customer's control panel.

Any number of system processes can cause the load to spike above 1.0 and having the light blinking red during that time alarms customers for no reason.

/me thinks Nick should think about changing this...

Originally posted by lightnin
Especially when it's accessible from the customer's control panel.
We have actually removed this from the CPanel template for our customers as it simply causes confusion and the help desk gets flooded with "Server High Load" tickets :)