Re: NFS Performance Woes

2014-07-22 Thread Ian Chapman

On 22/07/14 02:39, Peter Skensved wrote:


I've been using NFSv4 extensively for several years and I've not had an issue 
that you
describe where everything is fine and then suddenly performance goes to hell in 
a hand basket.

It sounds as if you only have 2 systems to work with?  No, tiebreaker so to 
speak?

Have you considered running a VM on your client system to see if it is affected 
in the same way?


   DNS problems can do it . Are your /etc/resolv.conf files correct ?
You could try running your own nameserver ( dnsmasq ) if the upstream
one is too slow or too busy.


I'm fairly sure it's not DNS. I run a DNS server actually on the same 
server, which serves NFS exports with the only DNS server in resolv.conf 
being itself (over localhost). All clients point to that DNS server and 
only that one. It's authoritative for my home LAN and both forward and 
reverse lookups work and resolve correctly and quickly too. Besides, the 
exports are specified by IP address on the server and the problem still 
occurs even why I mount an export from the client machine using the 
server's IP as opposed to its hostname.


--
Ian Chapman.
--
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


Re: NFS Performance Woes

2014-07-22 Thread Ian Chapman

On 21/07/14 21:27, Ed Greshko wrote:


For all intents and purpose it looks like its working as it should, it's just 
painfully slow.

Any NFS gurus out there, that can tell me what I'm doing wrong?


I've been using NFSv4 extensively for several years and I've not had an issue 
that you describe where everything is fine and then suddenly performance goes 
to hell in a hand basket.


Yeah it's really annoying. I could accept it, if performance was 
terrible all the time. I'd just figure I had some configuration setting 
out of whack, but it's the fact it works great.. then wham. To be honest 
it almost feels as if a cron job overnight it causing something to go 
screwy. It'll run fine all day, the next day, back to shit. That's 
without making any configuration changes.



It sounds as if you only have 2 systems to work with?  No, tiebreaker so to 
speak?


Actually, I have a couple of netbooks which hang off the wifi, although 
the NFS exports are set to read only on those, it's probably worth 
changing that and doing some further testing.



Have you considered running a VM on your client system to see if it is affected 
in the same way?


I hadn't considered that. Good idea. When the client's having a crisis, 
I can spin up the VM on the client and see if that's affected. I guess 
it would it least tell me if it's a hardware/driver quirk or a software 
issue.


--
Ian Chapman.
--
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


Re: NFS Performance Woes

2014-07-22 Thread Ian Chapman

On 21/07/14 20:14, Tom Horsley wrote:


Any NFS gurus out there, that can tell me what I'm doing wrong?


Not a useful answer, I'm afraid:

In my experience, the fundamental problem is caused by using NFS.

With all the folks re-writing things that don't need to be replaced,
I really wonder why no one seems to be re-writing NFS, which has been
utterly unreliable and trouble prone since day one :-(.


Sure, it has its issues, but to be honest most network file-systems do. 
CIFS is no better and in some ways takes more babysitting than NFS. 
Others don't have wide support amongst many platforms and then there's 
the differences in implementations. Do you use anything instead of NFS? 
I toyed with GlusterFS a few months ago and whilst it worked, following 
various snippets of info on the web, I found the documentation lacking 
and the admin tools confusing.


--
Ian Chapman.
--
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


NFS Performance Woes

2014-07-21 Thread Ian Chapman

Hi,

I've been unable to track down an extremely annoying performance issue 
with NFS. My home server has several exports shared over NFS4. The 
exports are all secured using Kerberos except one, which can be mounted 
using the traditional sys security model.


The problem is that write performance stalls at 48k/sec (or according to 
nfsiostat 4k/sec) despite being on a 1Gbit network and once this occurs, 
nothing I do seems to fix it. The only thing that does, is rebooting the 
client machine. This has been occurring for a year or so now but is 
happening so regularly now, it's driving me nuts! :-)


I'm fairly sure it's a client side issue, because rebooting the client 
allows normal transfer speeds once again. If I only reboot the server it 
makes no difference. My client is F19, the server is F20. After 
rebooting the client, write transfers can be fine for a day or so and 
then all of a sudden it hangs around 48k/sec again.


Restarting all the NFS daemons on the client makes no difference, 
including the associated RPC daemons. The NFS exports are mounted via 
autofs, but manually mounting makes no difference either.


When I'm experiencing slow writes to the server of 48k/sec, NFS reads 
are still in the region of 100MB/sec. Using scp or even copying files to 
the server via CIFS are well within the range of a 1Gbit network.


Nfsstat, wireshark and the system logs do not show anything which 
screams there's a problem.


The network card in the client machine and the server shows no 
collisions, dropped packets, frame overruns etc.


I've tested with the export that isn't using Kerberos and still have the 
same issue. Messing with the rsize, wsize, async, sync parameters makes 
no difference either.


The server has 32GB RAM, the client 16GB.

For all intents and purpose it looks like its working as it should, it's 
just painfully slow.


Any NFS gurus out there, that can tell me what I'm doing wrong?

--
Ian Chapman.
--
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


Re: NFS Performance Woes

2014-07-21 Thread Tom Horsley
On Mon, 21 Jul 2014 19:59:40 +0800
Ian Chapman wrote:

 Any NFS gurus out there, that can tell me what I'm doing wrong?

Not a useful answer, I'm afraid:

In my experience, the fundamental problem is caused by using NFS.

With all the folks re-writing things that don't need to be replaced,
I really wonder why no one seems to be re-writing NFS, which has been
utterly unreliable and trouble prone since day one :-(.
-- 
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


Re: NFS Performance Woes

2014-07-21 Thread Ed Greshko
On 07/21/14 19:59, Ian Chapman wrote:
 Nfsstat, wireshark and the system logs do not show anything which screams 
 there's a problem.

 The network card in the client machine and the server shows no collisions, 
 dropped packets, frame overruns etc.

 I've tested with the export that isn't using Kerberos and still have the same 
 issue. Messing with the rsize, wsize, async, sync parameters makes no 
 difference either.

 The server has 32GB RAM, the client 16GB.

 For all intents and purpose it looks like its working as it should, it's just 
 painfully slow.

 Any NFS gurus out there, that can tell me what I'm doing wrong? 

I've been using NFSv4 extensively for several years and I've not had an issue 
that you describe where everything is fine and then suddenly performance goes 
to hell in a hand basket. 

It sounds as if you only have 2 systems to work with?  No, tiebreaker so to 
speak?

Have you considered running a VM on your client system to see if it is affected 
in the same way?

-- 
If you can't laugh at yourself, others will gladly oblige.
-- 
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org


Re: NFS Performance Woes

2014-07-21 Thread Peter Skensved
 
 
 On 07/21/14 19:59, Ian Chapman wrote:
  Nfsstat, wireshark and the system logs do not show anything which screams 
  there's a problem.
 
  The network card in the client machine and the server shows no collisions, 
  dropped packets, frame overruns etc.
 
  I've tested with the export that isn't using Kerberos and still have the 
  same issue. Messing with the rsize, wsize, async, sync parameters makes no 
  difference either.
 
  The server has 32GB RAM, the client 16GB.
 
  For all intents and purpose it looks like its working as it should, it's 
  just painfully slow.
 
  Any NFS gurus out there, that can tell me what I'm doing wrong? 
 
 I've been using NFSv4 extensively for several years and I've not had an issue 
 that you
 describe where everything is fine and then suddenly performance goes to hell 
 in a hand basket. 
 
 It sounds as if you only have 2 systems to work with?  No, tiebreaker so to 
 speak?
 
 Have you considered running a VM on your client system to see if it is 
 affected in the same way?
 


  DNS problems can do it . Are your /etc/resolv.conf files correct ?
You could try running your own nameserver ( dnsmasq ) if the upstream
one is too slow or too busy.

peter
-- 
users mailing list
users@lists.fedoraproject.org
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org