RE: [SLE] how can I interprate this: cpu use is low, system is fine, loadaverage is high

在 2006-06-21三的 11:29 -0400，Marlier, Ian写道：

-----Original Message-----
From: 张|武 [mailto:zhangweiwu@xxxxxxxxxx]
Sent: Wednesday, June 21, 2006 8:44 AM
To: suse-linux-e@xxxxxxxx
Cc: wangpenghui@xxxxxxxxxx
Subject: [SLE] how can I interprate this: cpu use is low, system is fine,
loadaverage is high

Hello. I just logged in an X terminal server, noticed this X terminal
server has 3 users using it at the same time, one user is developing a
php application using Jedit (apache runs on anther host, he only use the
editor), I am writing email and browsing, the third guy is browsing the
web. The CPU usage is some 0% to 80%, avaragly looks like 10%. The
memory is used 94% (1GB ram), swap space almost not used at all, and
load average is 5.5

I learned early years in some articles saying load average is the number
of resource competiting for resource, a load average of 3 indicates the
system is under heavy load, a load average of 6 shows the system must be
upgraded or remove some service.

I don't understand why, when I can see the CPU usage is only 10%, the
load average says the system is under heavy burden. That doesn't make
sense. All the applications, evolution, OpenOffice, chat software,
Firefox... perform just like as fast as if I am the only one using the
computer. Is my feeling cheating me?

Or, can I interprate the load average of 5.5 as a OK value for X
terminal system? Does desktop application works in different way then
server application? Can this server handle more user by simply adding
more memory bank?

Remember that load average doesn't scale according to the number of cores available. A load of 3.0 on a machine with a single dual-core chip is just over half of the effective load of 3.0 on a machine with a single single-core chip. (Not exactly half because of throughput overhead.)

Load average is the average number of threads waiting at any given time. In the past, they were generally waiting for CPU. We've gotten to the point in CPU development, though, where it seems that waiting for I/O channel space is getting to be common.

I've seen load spikes (without associated CPU use jumps) on a number of machines lately; in each case, the root cause has turned out to be backups in the I/O channel.

Worth checking into, at least...

Since you mensioned this could be I/O blocking, this is my environment
and test data today:

1. 3 user use the X terminal server the same time;
2. average CPU usage is lower then 10%;
3. memory is used 95%, swap almost not used;
4. this is a single-core server;
5. I monitored HDD activity using gnome panel resource monitor, it
shows the HDD activity is very low, almost no activity
6. the system behave sanely and fast, opening openoffice less then
7 seconds (as it was with one user), doing everything thing
won't feel being slow
7. the only network service this machine have is 'svn', which is
used very in-frequently
8. load average currently is 8.5. It has been higher then 7 since
the early morning, now it's 17:00
9. I cannot understand 8).

Anyone please teach me something new:)

Is the 'load average' something really not for a dynamic system like X
terminal server? I mean, it can happen that when I open email software,
someone just hit 'gvim' and pressed enter, and two process compete for
CPU. This may only happen for 0.5 second, but that also counts. Such
accident can happen 100 times without user getting noticed, each time
lasts 0.5 seconds. and because it happened 100 times, it is counted for
the computer: "in last 5 minutes, process compete for resource 100
times, thus your computer must be under huge burden and let's give you a
surprising load-average"


--
Check the headers for your unsubscription address
For additional commands send e-mail to suse-linux-e-help@xxxxxxxx
Also check the archives at http://lists.suse.com
Please read the FAQs: suse-linux-e-faq@xxxxxxxx

Loading