Server load, and keyboards

by Juliet Kemp

I have a server (LDAP and NFS) which occasionally seems to take a while to react. Load average is consistently high (10-12 for 4 CPUs, which AIUI means 2.5-3 per CPU); response may also be being affected by disk I/O.

I'd like to find out which processes in particular are causing the heavy load, and possibly also to track disk I/O activity over say a 24 hr period. Unfortunately, this has exposed a shocking lack in my knowledge, viz: I have no idea how to do this, what tools are out there, etc etc. (Obviously I am fortunate never to have encountered performance issues before; all my machines have previously either Worked or Not Worked.) Any suggestions?

(The other possibility is a network load issue, but as I'm not responsible for the network, tracking that might be tougher.)

Another, unconnected query: I have been asked to source quiet/silent keyboards for a couple of my colleagues who are noisy typers. Any recommendations? I've looked at the Saitek Eclipse but it sounds like it's not very tough. Bonus virtual biscuit for recommendations which are actually available to buy in the UK (the IBM Quiet Touch appears not to be, for example). Real biscuit available to anyone able to implement a biscuit-over-IP protocol; failing that I shall just eat them all myself.


13 Comments

Tony Cossey
2008-03-13 07:26:25
Silent keyboards in the uk see this url http://www.clitheroelancs.co.uk/silent.htm
Istvan Belanszky
2008-03-13 07:31:44
Hi Juliet,


Another non-everlaster, but a silent one and a good finger joint saver is Logitech's Ultra-Flat keyboard. It has a small, slim form factor, and is very affordable.


For a bit less affordable, but apparently longer lasting, I can recommend any IBM keyboard that has the TPad-like slim, silent and flat keys -- I am using an SK-8835 multi-input combo device even for my Mac since August 2005, and it survived great in spite of some occasional MAME-ing and WASD-killer applications. \;]>


As to load profiling, whenever 'top' doesn't cut it, I use Activity Monitor.app. That thing is actually useful regardless of GUI eye-candy, and can do what you are looking for, but unfortunately it is OS X-exclusive.


Cheers,
Istvan

Casey
2008-03-13 07:53:40
Hullo,
A few good ways to look at this problem.
1) Try and install the python program 'iotop'
2) Look at 'sar' output. That will tell you how much IO you're pushing. Maybe one of your disks is being hit much more than the others, and you need to re-balance load or look for SCSI errors
3) ps auxw | egrep '\WD\W' will look for processes blocking on disk. Do this a few times a minute, append it to a file, and wait for an hour. You should have a pretty good idea of the processes you have to look at.
4) RRDtool. RRDtool. RRDtool.
Saint Aardvark
2008-03-13 08:34:15
Have a look at dstat. If it was a BSD machine, I'd suggest "systat -vm"...if anyone knows of a Linux equivalent, I'd be grateful for a pointer.
Russell Coker
2008-03-13 13:37:50
http://etbe.coker.com.au/2007/11/01/load-average/


You might want to read my above blog post.


For a server with two vastly different types of load (LDAP tends to be CPU bottlenecked on servers with any reasonable amount of RAM in my experience) the load average is not much use. Concentrate on CPU use as reported by top.


Also use "iostat -x 10" (or similar) to see which filesystems are being hit if it's disk IO (I'm assuming that you used different filesystems for NFS and LDAP).

José Luis Ledesma
2008-03-14 02:39:50
Use 'sar', 'vmstat', 'iostat' and which processes in the 'ps axuw' have STAT D
regards,
bob walker
2008-03-14 04:01:17
the options i can think of are:-


sar - http://packages.debian.org/stable/admin/sysstat
collectd - http://collectd.org/


i quite like dstat (http://dag.wieers.com/home-made/dstat/) for real time monitoring.


of course you could just use iostat, vmstat, netstat and the like or indeed top :)

ityokel
2008-03-14 05:28:33
I have found htop to be a little more informative than top, and colourful to boot.
Lindsay Holmwood
2008-03-14 19:10:32
I absolutely second Bob Walker's recommendations.


I spend a lot of my sysadmin time doing performance analysis, and the sar/collectd combination is a winner.

Charles Nadeau
2008-03-18 12:34:57
Juliet,


To monitor servers usage I use atop (http://www.atcomputing.nl/Tools/atop/) and dstat (http://dag.wieers.com/home-made/dstat/). Both of them can log to disk.
Cheers!


Charles

abitaniAdeday
2008-04-19 07:49:05
Хостинг:
Размещение в Европе - канал отличный, траффик безлимитный
Хостинг для doorways и не только (CP запрещено)
Разрешены любые виды прямого спама кроме e-mail спама
Лимит места на винте по договоренности;
Количество доменов/сабдоменов/БД неограничено
Plesk Panel or DirectAdmin
Цена за хостинг под doorways/redirects/images/shops/warez - от 20Э в месяц(зависит от места которое Вам нужно), цена за хостинг другого контента - по договоренности.
За информацией:
ICQ: 393513
e-mail: art.xeos@gmail.com
vaa guu
2008-06-18 03:40:33
NFS problem.


1. Check your /etc/export param
2. Check your nfs version


1 - change to ( rw,no_root_squash,async ) - I hope you have 'sync'
2 - Debian have two different nfs-server packeges
a. nfs-kernel-server - use this choice. ( native kernel mode ) If you using different linux distro find package when using native kernel mode.
b. nfs-user-server - runing in user mode, very slow and need more system resources.

vaa guu
2008-06-18 03:40:41
NFS problem.


1. Check your /etc/export param
2. Check your nfs version


1 - change to ( rw,no_root_squash,async ) - I hope you have 'sync'
2 - Debian have two different nfs-server packages
a. nfs-kernel-server - use this choice. ( native kernel mode ) If you using different linux distro find package when using native kernel mode.
b. nfs-user-server - runing in user mode, very slow and need more system resources.