[owncloud-user] Which is the best server configuration for ownCloud?

Pierre Malard plm at teledetection.fr
Tue Feb 25 21:27:50 MET 2014


Le 25 févr. 2014 à 18:07, Nicolas Zedde <Nicolas.Zedde at plug-it.com> a écrit :
> On 02/25/2014 4:52 PM, Pierre Malard wrote :
>> We have, since 2 years, an owncloud service on our institute network. Now,
>> we want to generalize its usage but we have some interrogation about the
>> best configuration.
>> 
>> The system works on virtual XEN server with Debian Wheezy. The test
>> configuration is this:
>> - 2 CPU (i7)
>> - RAM : 4096 Mo
>> - Storage area: NFS mont on a low-cast SAN SATA.
>> 
>> With about 5 or 6 users, the memory is alway to 90%. So what could be this
>> charge with 150 or 300 users?
>> 
>> is it because the IO access to NFS storage is too poor?
> 
> 
> Memory usage seems very high on your server compared to the number of users.
> Actually, I checked on an OwnCloud server on my own with roughly the same number of users and found 430MB for RAM Usage.
> Are you sure you're talking about used memory ? Debian uses cache memory to 90% of the system memory but it's not really used and can be allocated to applications.
> What does a 'top' command return in your case ? (you have to look at the ' -/+ buffer/cache' line)

That's ok, look the mem usage:
localu at nuage:~# free
             total       used       free     shared    buffers     cached
Mem:       4190044    3850664     339380          0     344996    1976800
-/+ buffers/cache:    1528868    2661176
Swap:      1048568       7260    1041308

I have increase memory allocated to this virtual machine (first test with 512 Mb) because we had a lot of strange error messages during synchronizations (not enought space left, …). As I saw the CPU usage were also full, I decide to gave 2 CPU. Now, the CPU usage is normal.

A explanation of memory usage with only 5 to 7 testings users could be that we have 2 ownCloud services: one for production usage and one for testing the new versions. The first one is on ownCloud v5.0.14a and the second on on last v6.0.1.

All ours user use this repository with all fonctionalities (data storage, calandar, …). The disk space allocated is from 500 Mb to 7 Gb per user. Every connections are done on our LDAP service.

When I'm take a look with "top", I can see that CPU usage is:
1st : apache
2nd : mysql
3rd : fail2ban

the memory usage top is:
1st : clamd
2nd : mysql
3rd : apache

If I talk about disk performance on my first mail, it's because the synchronization can be very hard and our data storage is, actualy, on a low Sata SAN. On the past, when we had only 1024 Mb of RAM, I have lots data during synchronizations following errors. I don't want ours users have the same problem.

So, my question can be:
If we have a lot of real users (about 50 concurents access to 300 users) which is the best configuration to our server?
- how many CPU dedicated?
- how many RAM?
- Sata disks or FC?

If someone have experience...

Regards

-- 
Pierre Malard

   « La liberté de la presse ne s'use que lorsqu'on ne s'en sert pas ! »
                                                  Slogan du "Canard enchaîné"
   |\      _,,,---,,_
   /,`.-'`'    -.  ;-;;,_
  |,4-  ) )-,_. ,\ (  `'-'
 '---''(_/--'  `-'\_)

perl -e '$_=q#: 3|\ 5-,3-3,2-: 3/,`.'"'"'`'"'"' 5-.  ;-;;,-:  |,A-  ) )-,_. ,\ (  `'"'"'-'"'"': '"'"'-3'"'"'2(-/--'"'"'  `-'"'"'\-): 22PLM::#;y#:#\n#;s#(\D)(\d+)#$1x$2#ge;print'
- --> Ce message n’engage que son auteur <--

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 841 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://mailman.owncloud.org/pipermail/user/attachments/20140225/6d42bf6b/attachment.sig>


More information about the User mailing list