Networker group performance question

Question

Hi,

I have the folowing folder structure:
/app folder with 3 subfolders ./ap1 ./ap2 ./ap3

subfolder ./ap1 has ~ 600.000 files and 12 GB
subfolder ./ap2 has ~ 1.200.000 files and 28 GB
subfolder ./ap3 has ~ 50.000 files and 5 GB

What is the best way to configure the backup process for the folder /app?
1. Should be a single group with a single save set /app (does networker use in this case multiple streams) ?

2. Should be a single group with the save set containing
/app/ap1
/app/ap2
/app/ap3
Does networker better manage the save streams in this case

3. Or should be 3 different groups each oane containing the specified folder
one group for /app/ap1 one for /app/ap2 and one for /app/app3 ?

Which one is better for performance ?

Thank you for any suggestions.

ble1 · Answer

It depends on how much your disk can deliver. The best way is to test it (2&3 are almost identical meaning while there is difference you can achieve the same thing with both approaches).

dpinink_silva · Answer

As long as your environment can support that, open as many savestreams as you can. You can try first option and then second option and see which one gives you the best result.

If the first option gives you better results, you may have a bottleneck somewhere in your environment that you should improve (can be disk, network speed, drive speed etc etc).

Performance is very specific for each datazone, and the best suggestion is: try, try and try different ways.

ble1 · Answer

It depends on your configuration (eg. parallelism on client and group).

razvan2 · Answer

You are right, I should try every case but I don't want to think that networker is a black box and you must try to find what's happen.
I would like to understand how networker treats these cases in the same conditions. How it decides the number of streams.
For example if I will start 3 separate groups (each folder in its group) at 1 min interval, all writing to a disk pool or 1 group containing all these folders also to the same disk pool is there a theory that tells me how networker will treat these cases.
Of course I will test it but I must know if I can anticipate the result based on networker documentation.

razvan2 · Answer

Client parallelism, quote from networker help:
"Parallelism - The number of save sets to run in paralle". In my case this is = 1
But in one case I'm running a single save set that contains 3 folders, so this parameter does not have any influence.

In the second case I have 3 groups everyone with its save set so 3 save sets. If this parameter remain 1 I understand that each group will stay in line waiting the other groups to finish. If this parameter will be=3 I will have 3 parallel streams.

If we presume the same conditions the second case should be a faster backup.
Is this correct ?

I also understand that a "Parallelism" of let say 5 for a client with a single save set does not have any effect. Is this correct?

thanks

ble1 · Answer

But in one case I'm running a single save set that
contains 3 folders, so this parameter does not have
any influence.

Correct.

In the second case I have 3 groups everyone with its
save set so 3 save sets. If this parameter remain 1 I
understand that each group will stay in line waiting
the other groups to finish. If this parameter will
be=3 I will have 3 parallel streams.

In second case you could have 1 group also with 3 savesets and in that case p=1 on client would be the same as 3 groups with each saveset starting one after another. That is the same as having first case, but you make data volume that you backup smaller.

That parameter has nothing to do with groups - if you want them to wait you should go for one group with client parallelism 1 or savegrp parallelism 1 (only if that would be only client in the group).

If we presume the same conditions the second case
should be a faster backup.

I think they will be the same become t(big volume)=t(small1)+t(small2)+t(small3).

What you must test and see is can you have parallel backups against same disk (eg. small1 and small2 or even all 3) without decreasing disk read performance (and machine too).

I also understand that a "Parallelism" of let say 5
for a client with a single save set does not have
any effect. Is this correct?

Correct.

NetWorker

Networker group performance question

Was this post helpful?