Fine Tuning Linux for backup/production coexistence.

Hi,

I am running a CentOS 4 box with 4 terabytes of data on it. It houses
most of my users' data, so it is constantly being used.


I'm running the Networker 7.3.2 client on the box, as well. I have
established a seperate backup network successfully - all data is
backed up over this secondary network, and not over the production
network. The secondary, like the primary, is all GigE, so we're as
fast as we can go, there.


Even though I've made this change, machine response is VERY slow
during a backup. The CPU(s) get thrashed, since it has a lot of
calculations to do in order to perform a backup.

Sadly, there are no throttling options in Legato Networker, since its
really designed to do a backup as quick as possible. What I'd like to
be able to do is a backup of the system, without (significantly)
impacting server performance.

Now granted, I try to run backups during non business hours, but
sometimes thats imposssible. I tried to /bin/nice the save processes
on the file server, but this did not have a significant change.

Is it possible to improve the performance (without completely swapping
out the back plane?) Are there any /proc settings that might help?
Would bumping up the number of NFS processes help? How about the wsize/
rsize of the exported directories? Anything else anyone can think of?

If anyone has any thoughts, I would really appreciate it. I'm at a
loss, and I don't even see the advantage to a backup network now,
since my network is STILL impacted when a backup is running.



- Thanks,


Todd

"Todd" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed) oups.com...
> Hi,
>
> I am running a CentOS 4 box with 4 terabytes of data on it. It houses
> most of my users' data, so it is constantly being used.
>
>
> I'm running the Networker 7.3.2 client on the box, as well. I have
> established a seperate backup network successfully - all data is
> backed up over this secondary network, and not over the production
> network. The secondary, like the primary, is all GigE, so we're as
> fast as we can go, there.
>
>
> Even though I've made this change, machine response is VERY slow
> during a backup. The CPU(s) get thrashed, since it has a lot of
> calculations to do in order to perform a backup.
>

GigE is good, but GigE w/ TOE is better. This will offload the processor.
If the backup application is the culprit, there's nothing you
can do except insure that your system has adequate memory. Monitor the page
faults if this is possible. If it's excessive, more RAM may resolve the
sluggishness.

Da Moojit

> Sadly, there are no throttling options in Legato Networker, since its
> really designed to do a backup as quick as possible. What I'd like to
> be able to do is a backup of the system, without (significantly)
> impacting server performance.
>
> Now granted, I try to run backups during non business hours, but
> sometimes thats imposssible. I tried to /bin/nice the save processes
> on the file server, but this did not have a significant change.
>
> Is it possible to improve the performance (without completely swapping
> out the back plane?) Are there any /proc settings that might help?
> Would bumping up the number of NFS processes help? How about the wsize/
> rsize of the exported directories? Anything else anyone can think of?
>
> If anyone has any thoughts, I would really appreciate it. I'm at a
> loss, and I don't even see the advantage to a backup network now,
> since my network is STILL impacted when a backup is running.
>
>
>
> - Thanks,
>
>
> Todd
>

In article <45c4e024$0$5236$(E-Mail Removed)>,
Moojit <(E-Mail Removed)> wrote:
>
>"Todd" <(E-Mail Removed)> wrote in message
>news:(E-Mail Removed) roups.com...
>> Hi,
>>
>> I am running a CentOS 4 box with 4 terabytes of data on it. It houses
>> most of my users' data, so it is constantly being used.
>>
>>
>> I'm running the Networker 7.3.2 client on the box, as well. I have
>> established a seperate backup network successfully - all data is
>> backed up over this secondary network, and not over the production
>> network. The secondary, like the primary, is all GigE, so we're as
>> fast as we can go, there.
>>
>>
>> Even though I've made this change, machine response is VERY slow
>> during a backup. The CPU(s) get thrashed, since it has a lot of
>> calculations to do in order to perform a backup.
>>
>
>GigE is good, but GigE w/ TOE is better. This will offload the processor.

Do you have any rational reason to believe that a gigabit of TCP
throughput will saturate a modern processor?

The dirty little secret about "TOE" is that it often _slows things down_.

--
Thor Lancelot Simon (E-Mail Removed)
"All of my opinions are consistent, but I cannot present them all
at once." -Jean-Jacques Rousseau, On The Social Contract