Hi everybody. This are outputs of the command "uptime" on my shared ("virtualized", if we want to be precise) server, a Freedom 200 by Hostpro. The site is responding, but aren't these too much???
I alerted many times the technical support, but they always reply "we have told technicians to monitor your account and the server on which you are on"
What do you suggest? Anybody else in my situation?

> uptime
2:59PM up 21 days, 4:39, 1 user, load averages: 10.46, 12.98, 11.14
> uptime
3:55PM up 22 days, 5:35, 1 user, load averages: 11.54, 8.14, 6.54
> uptime
12:35PM up 55 days, 2:25, 1 user, load averages: 13.94, 13.62, 11.03

Thanks in advance.
:angry: :angry: :angry: :angry: :angry: :angry: :angry: :angry: :angry:

Originally posted by marco
> uptime
2:59PM up 21 days, 4:39, 1 user, load averages: 10.46, 12.98, 11.14
> uptime
3:55PM up 22 days, 5:35, 1 user, load averages: 11.54, 8.14, 6.54
> uptime
12:35PM up 55 days, 2:25, 1 user, load averages: 13.94, 13.62, 11.03

Those are QUITE high. What OS are they running? Try checking it early in the morning and late at night as well.

How does the server perform? Is it still fast? Does it ever "lock up" and then unlock itself and continue serving? It's obviously good at staying up, hence the days up. Run "top" and see what you find. Is the memory almost completely being used? Are there an insane amount of processes? Can you see what processes are eating up CPU?

You might want to ask to be moved to another server with a less load on it. :)

Thanks for replying. Running "ps fuax" and "top" doesn't tell me nothing because I will see only MY OWN processes (the account is shared; thought, it acts as if you had a dedicated server on your control)
The same happens for the used/unused memory and disk quota.

Here you can see some system specs (I think that this FreeBSD version has a *VERY* deeply modified kernel, BTW):

> uname -a
FreeBSD uffa.it RELENG_4_2001_03_07 FreeBSD RELENG_4_2001_03_07 #0: Thu Mar 8 02:08:40 GMT 2001 eanderso@osbuilder:/usr/src/sys/compile/SERVER i386

Any help? I think that the server has 2 CPUs (don't know which)
Is there anybody hosted with a Freedom account at Hostpro and has my same problem???
:bawling: :bawling: :bawling: :bawling: :bawling: :bawling:

Dood. Thats terrible.. Type

"top"

and see what it constantly at the top of the list. It could be a program gone crazy. 10.xx = usually means that the server is pretty close to crashing. Your server load should try to stay under 1.00 but 2.00-4.00 is somewhat acceptible for high peaks.

Sorry to sound THICK but these three numbers are CPU load, i know that already, but i "thought" they were "percentage" load and that 10.xx is 10% loaded, now forgive me if im daft but id have thought 10% was pretty good for a busy server??

Actually.. I am pretty sure (almost 100%) it works like


.10 = 10%
10.00 = 1000%


I know you can't have 100%+ of use so it just saves the data waiting to be used which slows the hell out of the box.

If I remember correctly, load averages are the number of processes waiting for CPU time. So on a multi-processor system, 10.0 is equivelant to 5.0 on a single-proc machine.

No xstrike, a line like:
10.46, 12.98, 11.14

means that there were more or less 10 processes ready to run in the last minute, 12.98 in the last 5 minutes and 11.14 in the last 15 minutes

So, a load of 10 means that you are using the server at 1000% of his capacity.

Try running a fork() bomb (better with a malloc() call :D) on your linux box at home... you will see the /proc/loadavg going at incredible numbers... even with only 64 Megs of RAM

:)

P.S. jic: if I run top I get only my processes! The environment is virtualized

I remember having a line like:

3.00 2.45 3.50

and ive also had stats like:

0.00 0.00 0.01

its quite confusing as to what it all means, what would the above statistics mean for a box?

A line like 0.00 0.00 0.01 means that your server isn't overloaded at all. And that's what I was used to get on my server until one month ago, when the situation I have explained came, all of a sudden.

For instance, on usw-pr-vhost.sourceforge.net they currently have this:
8:06am up 8 days, 19:57, 24 users, load average: 0.32, 1.18, 0.85

The server, as you can see, is not loaded (and there are 24 users connected!)

Bye,

what i was thinking is there must be a mathematical scale used to calculate the load, as opposed to saying simply

"hmmm... let me think... yeah that looks light"

as people have said, it doesnt seem to be a scale of 0-100 like NT/2000

anybody know anything further about this?

As far as I know .. those numbers are Average time it takes to load a waiting process.
The first number is the Average per minute.
Second number is Average every 5 mintues
last number is average every 15 mintues

0-2 Light
2-4 Med
4-6 High
6+ over loaded.

The number you show at first say that over 1 mintues on average it takes 12 seconds to run a process.

That is very high.


I've seen servers as high as 35.00 .. then they always seem to crash :)

I more or less agree with Jason_Berresford's chart, though I might bump the numbers down a bit...

To me, 200% wouldn't fall into the 'light' catagory but maybe something more like...

0-1 Light
1-3 Med
3-6 High
6+ over loaded.

It isn't that high is bad exactly, just that if it is running that high most of the time, then a surge will certainly push it beyond what you'd want.

This is one of those things I've found that no one agrees on 100%. There is no exact rule, far as I can tell. Simply a server at 7.00+ won't be able to handle things as well, nor as fast as the same server at 1.00. Like ram (amounts of it), it is somewhat up to the person running the server as to what performance is 'good enough'.

the numbers corresponding to CPU Load actually have been discussed here quite a while back ... The three number you see when you type "uptime" are the CPU Load average for the last 1 minute, for the last 5 minute, and for the last 15 minute correspondingly.... As to what each number actually mean, it's just telling you how much capacity is the server currently running at, defined by this formula:

Let this number (CPU load) = x, then

CPU Capacity (in percentage) = x/(x+1)

Hence, a CPU Load of 0.00 means the CPU is running at 0% capacity (in other hand, the CPU is idle). A CPU Load of 1.00 means the CPU is running at 50% capacity, and a CPU Load of 10 means the CPU is running at 91% capacity. The higher the number of CPU Load, the busier the CPU is (hence, if you do "top", the CPU Idle number will be correspondingly smaller).

BTW, it does not make sense to me a this CPU Load will be interpreted as a CPU running more than 100% of its capacity (I see some numbers like 200% and even 1000% on this thread), since the max the CPU can run is at its full capcity, i.e. 100% ;)

cheers,
:beer:

Hence, a CPU Load of 0.00 means the CPU is running at 0% capacity (in other hand, the CPU is idle). A CPU Load of 1.00 means the CPU is running at 50% capacity, and a CPU Load of 10 means the CPU is running at 91% capacity. The higher the number of CPU Load, the busier the CPU is (hence, if you do "top", the CPU Idle number will be correspondingly smaller).

With the number I have seen that doesn't make 100% sence. For example .. here is stats from one of our servers.

1:23pm up 18 days, 17:53, 1 user, load average: 2.74, 1.00, 0.42

I placed one of our none hosting servers under a little bit of processing.

As you can see, the first number is 2.74, under what you are saying the server load would be 140% or so .. which is impossible.
I've seen that number at 30.0 and higher .. (which I might say made the server run .. very badly)

This would make sence if the load was actually telling you how many seconds it takes to run a processes. and not the % of CPU used.

Originally posted by Jason_Berresford
As you can see, the first number is 2.74, under what you are saying the server load would be 140% or so .. which is impossible.

Using the formula dektong used, which is what I believe you are referring to, a 2.74 load would result in a 73.2%, not 140%.

2.74/(2.74+1) = .732 or 73.2%

Don't know if there is any truth to that formula or not, but his formula does not result in 140%.

My mistake then :) .. However I have still seen server loads much higher then that. On a number of times.

Originally posted by Jason_Berresford
My mistake then :) .. However I have still seen server loads much higher then that. On a number of times.

No matter how high the number goes, using the formula that I quoted (from somebody else, actually ... a long ago post here), then it won't never go higher than 100%, since x/(x+1) is always less than 1 (or 100%) :) ... Hate to see those numbers go beyond 1 though ;)

cheers,
:beer:

Yep ... *hits head* Your right.

I've said it before and I'll say it again, I've seen a server with a load of 90+ (It was dual CPU so effectively 45+) and it was serving pages and doing its tasks at its normal speed.

In case you were curious the server was targeted for a DoS attack.

Aloha
well I have had good luck with vserver(hostpro) for over 4 years and am now jumping ship there win2k support is very sad they take days now to do most anything and are getting worse all the time so sad
they have very good uptime though and fairly good speeds but have also noticed that there machines are always under a high load
the other thing is that the lower machines have at least 1000 accounts on them
after 4 years I am weening myself off hostpro still looking for someone but have not been happy so far with anyhosts so might be going my own servers and colo

good luck Marco

I think that formula came from me; it isn't exactly correct and it makes some wildly inaccurate assumptions about server processes (namely, that they wake up randomly and independantly), but it is good enough to act as a rule of thumb.

If you want more accurate numbers, try `vmstat -w 2` and look at the last number on the line (the idle cpu %). Subtract that from 100 and you'll get the cpu utilization.

(Load Average is really rather misnamed; a better term for it would be something like "responsiveness average" since it indicates how long processes wait before being given cpu time.)

I hate to play one-up-manship but I saw a server with a load over 500.. granted it was just before it crashed... but I did see it....:D

damn, no wonder people that arent rocket scientists dont want to use linux, i mean lets face it, if you cant even get the cpu stats without going to the moon and back, what hope is there for the rest of the operating system imho, lol

I think the formulas have cleared alot of things up though, and I will try vmstat when i get back home tonight, I will tell you how it goes.

Start TOP and exit it via Control-Z (NOT q). De day after you will see a server load of 1.5 (Yes, over 1).

It happened to me and TOP was using 99% of the available 256mb memory and was going berserk on the server.

Imagine what happens when other users don't exit it properly the right way.

A cron job that kills the TOP command twice a day or so will do the trick I guess.