NET_BUFFER

[mb_jedsv]

We have slowness in NetBackup 6.5 in peak time on some clients not all, and I found this solution to help us fixing that issue:

http://www.symantec.com/business/support/index?page=content&id=TECH1724

 I configured TCP/IP NETWORK BUFF SIZE to MASTR-SRVR, but the performance still the same, and I gana to change DATA BUFF Size & NUM of DATA BUFF.

But we are using the master server as a media server, and the DATA BUFF Size & NUM of DATA BUFF are meant exclusively for media and should not be used on a pure master server or client.

Is there any problem if I apply this step direct to Master Server?

Are we working on the right path?

Do we need to stop and start all NetBackup services after these changes?

[Marianne]

Is the Master also the Media server?  Please make changes to media server and clients as per the TN:

"Specifying a larger value in the NET_BUFFER_SZ file may improve the performance of backups and restores. It is recommended that if NET_BUFFER_SZ is used, the same value should be set on all the NetBackup media servers and clients."

Please note that the change "may improve the performance of backups" . Create bptm log on media server to study effect of change before and after. Please have a look at Planning and Performance Guide  http://www.symantec.com/docs/TECH62317  for guidelines on how to interpret backup results in bptm log and how to make adjustments based on results.

No need to restart NBU for changes to take effect..

[Yasuhisa_Ishika]

If performance degration is came from buffer, all client must be affected. And, I have heard from Symantec Principal Engineer that, NET_BUFFER_SZ does not need to be changed in most case(almost all, he said). Usually, performance degration is made in client side(many small files, fragmentation,...).

[mb_jedsv]

 I did Marianne’s recommendation and I got good performance in off peak time, so I think Yasuhisa Ishikawa is right we have a lot of small files on that backup, hence I think we will reschedule some backup police to be active in off peak time.
 

But do you think changing “DATA BUFF Size & NUM of DATA BUFF” will do any changes in performance?
We use  “Master Server as a Media Server”

Thanks for all
Regard

[Yasuhisa_Ishika]

Changing these parameters are really effective. You can get up to 100MB/s(or so) transfer per stream with default value in NetBackup 6.x or later. This is enough for LTO Gen 2, but if your tape drives(or disks) is faster, you should change these parameters to get better *possible* performance. # if performance bottleneck is not in buffer processing, buffer tuning does not take effect inperformance. Wait and delay counter detailed in Backup Planning and Performance Tuniing guide is useful to determine whether bottleneck reside in storage side or in client/network side. If counts in both side(waiting for empty/full buffer) are high, you should set grater value for these parameters. Be sure not to set excessive value as SIZE_DATA_BUFFERS. This lead to backup or restore failure. I prefer to use default value for SIZE_DATA_BUFFERS, and tune NUMBER_DATA_BUFFERS.

[mb_jedsv]

is that okay to apply "SIZE_DATA_BUFFERS & NUMBER_DATA_BUFFERS ?

[Yasuhisa_Ishika]

No problem!

[David_McMullin]

Be aware that your SIZE BUFFER and NUM BUFFER will use your memory - you will need to calculate your estimated maximum concurrent jobs and decide how large to make them. You can run your server out of memory and jobs will fail if you have none to allocate to buffers.

ALL LOCAL DISKS and multiple streams will multiply the job count - factor them in!

[mb_jedsv]

Oh … thanks a lot for this point, it wasn’t in my calculation.

Could you please help me to do this calculation? Or is there any document for this?

Regard

[Marianne]

Have you had a look at Planning and Performance Guide  http://www.symantec.com/docs/TECH62317  yet?

extract:

Amount of shared memory required by NetBackup
You can use this formula to calculate the amount of shared memory that NetBackup requires:

Shared memory required = (number_data_buffers * size_data_buffers) * number_tape_drives *  max_multiplexing_setting

For example, assume that the number of shared data buffers is 16, and the size
of the shared data buffers is 64 kilobytes. Also assume two tape drives, and a
maximum multiplexing setting of four. Following the formula, NetBackup requires
8 MB of shared memory:

(16 * 65536) * 2 * 4 = 8 MB

[mb_jedsv]

Thanks a lot.

I will do it. and i will write the result.

[mb_jedsv]

we have 30 data buffers & 64 kb data buffer size:

12:00:32.411 [25434] <2> io_init: bpbrm_shm_id = 50331702, buffer address = 0xfdb80000
12:00:32.411 [25434] <2> io_init: found index: 0
12:00:32.411 [25434] <2> io_init: using 65536 data buffer size
12:00:32.860 [25434] <2> io_init: CINDEX 0, sched Kbytes for monitoring = 40000
12:00:32.861 [25434] <2> io_init: using 30 data buffers
12:00:32.861 [25434] <2> io_init: child delay = 10, parent delay = 15 (milliseconds)
12:04:07.781 [25782] <2> io_init: bpbrm_shm_id = 56, buffer address = 0xfdb80000
12:04:07.781 [25782] <2> io_init: found index: 0
12:04:07.782 [25782] <2> io_init: using 65536 data buffer size

shuld i increase data buffer size to 256 kb , and leave the number of buffer as is "30"?