Web Hosting Talk : Web Hosting Main Forums : Dedicated Server : how high can the load average go before..

[disgust]

you figure you need to stop adding sites/users to your server and move to a new one?

and assuming it's reaching that point, what do you try to do to see if you can optimize it at all?

(this is on a Celeron 2.4ghz by the way)

edit: should read load average, not user load

[amusive.com]

Typically when load average equals the number of physical processors in your machine.

[BitError]

Premium hosts consider any peak above 1.0 on the 15 minute unacceptable. Multiple processor machines further mitigate the damage a single run-away process can do to the overall function of a machine.

Budget hosts tend to add users until http responses are very noticably delayed, at which point they shutdown high usage accounts or shut off their phones.

[f0urtyfive]

I remember I saw an article on how load averages were calculated a while back... Just to let you know, 1) it was not something that made very much sense (it was inredibly complicated) and 2) load averages arent a "measure", they arent the same on all machines. For example I could have a load average of 30 on my server and it could still be operating fine, while you could have a load average of 2 and it could be trashed.

[geeks4help]

If it is a single processor, any load above 1.0 for a prolonged period is a high load.

[Dave - Just199]

This is one of my core web servers:
load average: 0.46, 0.42, 0.27

At this point I am currently looking to upgrade this server.

I try to keep all of my Standard Web servers below .5

I have a server specifically for e-commerce and it can run higher but the machine is geared towards this type of usage and it does not slow down too bad 3.5

on a celeron you should be upgrading the server before 1.0 in my opinion.

One of my core web servers that i consider 25% full runs at:
0.02, 0.03, 0.03

[sprintserve]

Any discussion on load average is meaningless without looking at the system it is running on. Without that information, it's just a 'number'

You must also see if there's a particular event that may be unavoidable or whether it's a sustained load level. For example nightly backups. During this period, our single CPU machines can hit a load of 5 easily. But it's for a short period, and you can schedule it for off-peak times (example after 4AM)

For a Celeron 2.4, basically you should keep the sustained load below 1 for it to run without noticable delays. The calculation of load figures already take in consideration (somewhat) the type of processor you have. It measure how many processes are in the CPU queue to be processed. So if you have a slower processor, the queue will be backed up more compared to a faster one. i.e. if you are running the same operation, since a faster dual machine can process it faster, their queue will be a lot shorter. So the load number will already be lower even if both machines are processing the same operation. Obviously many factors can affect it example memory or how fast your disk can read. More memory means data can be read faster (and thus sent through the queue faster), or if a process is reliant on a read from the disk drive, even if the CPU can run faster, the queue can still start accumulating. (That's why during I/O intensive task, your load will increase too)

When we used to run single CPU machines as our primary machines a year or so ago, we try to make sure we are always below 1 most times, and we started to scout for a machine once it sustain at 0.5.

It should also be noted that noticable difference is also different from dependent on the speed of the cpu. This is natural because a single slower processor for example isn't going to take the same time as a faster dual processor machine in processing 1 process assuming it's the same operation.

Hope this explains.