How to optimize drives usage?

[picapiedra]

Hi all, and thanks in advance.

Scenario:

NBU 4.5 FP5

1 Master/Media Server, Solaris.
1 Media Server, Solaris.
25 Solaris clients.
25 Windows 2003 clients.
SAN environment for some Solaris/Windows clients.
Devices: both are SUN L60, so I have twelve DLT8000 drives.

I´m trying to backup one W2003 server with 150 Gb, and I can use up to six drives for this policy.

How should I set the policy to use more than one drive, reducing time? I know I´m going to need more than two tapes, and I don´t want this backup to take 8 hours to finish, so, why not use three drives at the same time?
I know about the *.utl file in the Solaris environment, but, what about the W2003 clients?

Any help?
Thanks again.

 

[comtec17]

How are your storage units configured?

How is the Win2003 disk configured? Is it 1 array or multiple?

Is the 2003 server on the SAN?

Bad thing with using 3 tapes or 3 drives is that your restore will take forever. When planning for DR, you should consider your restores times first, then backup times. It is all fine and dandy when your backups are done in 3 - 4 hours but it sucks when it takes you 2 days to recover a host.

 

[picapiedra]

Thanks for such a quick reply, comtec17.

I have two SUN l60 libraries, six DLT8000 each one, both are configured as follows:

Max Concurrent Drives=6, on demand, Max Fragment Size=unlimited, Max MPX=1.

W2003 server is on the SAN, any clue?

Tell me if need more info, I´m working till late, because of the new thread I´ve just opened about VSP with W2003 (please read it).

 

[comtec17]

Are you multiplexing your backups?

If so,set your multiplex level to 6 in your policy and schedule, then break your 150GB drive into 6 multistreams in the files attribute. Example below:

NEW_STREAM
E:\[a-e]*
NEW_STREAM
E:\[f-n]*
NEW_STREAM
E:\[o-s]*
NEW_STREAM
E:\[t-z]*
NEW_STREAM
E:\[0-5]*
NEW_STREAM
E:\[6-9]*

This is just an example....

 

[picapiedra]

Thanks, comctec17. I´m going to test it.

So, for W2003 clients, there is NOT any other way to use more than one drive in the same policy?

If this is correct, I have a new problem right now.

I have to backup six W2003 clients every night. This number is increasing as we are migrating server from NT(ARCserve) to W2003(Netbackup). I could use up to 10 or 12 drives, but I only have one policy for all servers, because the configuration (except the SQL server and Domino when we get the agents) is the same for all of them. In the files tab I have ALL_LOCAL_DRIVES and SYSTEM_STATE (Veritas recommendation for IDR), but nothing else (no strems defined). I´m not multiplexing my backups, but the 150 Gb Domino server took yesterday up to 7 hours to finish (Domino stopped), using one drive (three tapes). I was thinking to save this backup as a special one (post-migration) and add today this client to my W2003 only policy, but, what about using six, eight, or ten drives to reduces backup time? My final scenario is having 26 W2003 clients, two L60 libraries (12 drives), and a 7 hours daily backup window. And..?

Is there any easier way to backup all my W2003 servers (ALL_LOCAL_DRIVES and SYSTEM_STATE) in parallel using six or ten drives, than creating different policies, and defining streams in the files tab?

Thanks again, waiting for any help.

Picapiedra.

 

[comtec17]

If you check the use multiple drives in the Attributes of the policy and change your streams in the schedule to 6 you will be able to use:

ALL_LOCAL_DRIVES
SYSTEM_STATE

You will get 1 job for each drive and one for the System state.

 

[picapiedra]

I understand what you say, Auto-discover multistreaming, isn´t? but...

I don´t have any "Use multiple drives" box to check in the Attributes Tab.
And..where do you change the streams in the schedule?

I´m missing something, I think...

If I check "Allow multiple data streams" in the W2003 policy Attributes Tab, with

ALL_LOCAL_DRIVES
SYSTEM_STATE

in the directive list, Auto-discover will generate one stream/job for each drive and one stream/job for the System State, but I have 3 drives for client, so... 4 streams X 26 clients = 104 backup jobs everynight?
The fact is that I don´t really care about most of them, because there are lots of 2 or 3 GB drives (C:\ or D:\..) but there are some drives in some clients with 130 or 140 GB. These drives are the conflictive, (these drives are on the SAN, so streaming them will not adversely
affect backup times, I think), because this drives streams would take several hours to finish...
How to optimize all that (always having present IDRs...)??

I´m thinking about creating new policies for this drives and manually configure the directive list for Administrator_Defined streaming... (ufff) and...I should exclude this drives in the main W2003 policy, shouldn´t I?

Also, my main policy runs one job per client, so, limitating max number of jobs per policy in the Attributes Tab I can control the parallelism and drives usage...

 

[comtec17]

Do you have to do full backups every night? Can you not do differential incremental?

 

[picapiedra]

Unfortunately not.
The bussiness say, I do.
I´m trying to introduce a change in some minds, but, you know... Full is easier/faster to recover? Lets full. Don´t care about money/administration/operation...
I would like not to do fulls, but...

 

[PGPhantom]

Why not cumulative incrementals - Saves a lot of tapes, only two tapes needed for restore (2 images, possibly moretapes if large). Quicker, easier.

 

[picapiedra]

I´m sorry, but unfortunately this is not an option. (Not at this moment, believe me, this is a "pending request" for me...)
Full backup of every server everyday.

 

[picapiedra]

Bad news (at least for me..)

I tested last night creating a new policy for the 150 Gb folder, enabling multistreaming, and creating three streams, so, in the files tab:

NEW_STREAM
E:\Lotus\Domino\Data\mail\[a-j]*
NEW_STREAM
E:\Lotus\Domino\Data\mail\[k-t]*
NEW_STREAM
E:\Lotus\Domino\Data\mail\[u-z]*

In the W2003 "main" policy I added this client, but excluded the path E:\Lotus\Domino\Data\mail\* in the client properties for this policy and all schedules.

What happened?

The main policy was ok, backing up every W2003, including the Domino one, but excluding the path E:\Lotus\Domino\Data\mail\* for it.

The special policy created for the 150 Gb path E:\Lotus\Domino\Data\mail
started three jobs, but all finished in a few seconds with Status Code: 71

Message: none of the files in the file list exist

Explanation: The files in the file list did not match any of the files on the client.
This error can occur when there is only one file in the file list and the file
cannot be backed up due to an I/O error.

I´ve checked the recommended actions, but I was unable to discover anything. Is there any syntax error?

I had window enough to backup the whole path in the same job, so I created a copy of this policy with

E:\Lotus\Domino\Data\mail\*

and it has finished OK, (6 hours).

But I want to reduce time using three drives, I don´t want to resign to it.

 

[PGPhantom]

I am assuming that you have your policy set up to allow multi streaming? When you try the backup - What is recorded in the log files? Specifically the bpbrm and bpcd logs?

 

[picapiedra]

Yes, multistreaming is allowed. I´ve activated logging for tonight backup, but I´m thinking this is just a syntx problem. I was assuming (according to the product documentation), that "...You can use either Microsoft Windows conventions or UNIX file-path conventions,
whichever you are the most comfortable with. You can also mix the two styles within the same file list."
But I think this is not exactly correct.
I´m going to test tonight with Windows conventions:

NEW_STREAM
E:\Lotus\Domino\Data\mail\a*
E:\Lotus\Domino\Data\mail\b*
....
E:\Lotus\Domino\Data\mail\j*
NEW_STREAM
E:\Lotus\Domino\Data\mail\k*
E:\Lotus\Domino\Data\mail\l*
...
E:\Lotus\Domino\Data\mail\t*
NEW_STREAM
E:\Lotus\Domino\Data\mail\u*
E:\Lotus\Domino\Data\mail\v*
...
E:\Lotus\Domino\Data\mail\z*

Tell something tomorrow, but anyway, does anyone know any more correct way to define this paths/ file list?

 

[PGPhantom]

I should of read more carefully - I thought you were doing a UNIX backup as you are correct in your latter assumption - Windows, as far as I know and have tested - Does not know what [a-j]* etc means.

Just remember though - For every stream that you have, the bpstart and bpend scripts will be run.

 

[enterprisebu]

I recently halved my backup times by tweaking the buffer sizes on clients, media servers and master. I started with master and waited, then changed the medias and waited and finally modified the clients. I got performance improvements each step of the way. Wish I'd just done them all at once. One particular media server used to take 13 hours for a full and last weekend finished in 4.5 hours!

 

[PGPhantom]

I can believe it - Those buffers are all too important.

 

[mazrimt]

enterprisebu, any site in particular you used as a guide to the tweaks? If not could you post what you did?

 

[comtec17]

Go to

http://support.veritas.com

and download the performance tuning guide. This guide has all the buffer changes and much, much more.

 

[enterprisebu]

This thread gave me information for the windows and I'd read it in the 5.0 documentation, I believe.

http://www.tek-tips.com/gfaqs.cfm/lev2/7/lev3/49/pid/776/fid/3124