Video Screencast Help
Symantec to Separate Into Two Focused, Industry-Leading Technology Companies. Learn more.

Netbackup Disk Performance & Caching on Windows

Created: 01 Nov 2012 • Updated: 02 Nov 2012 | 5 comments

Hi

I’m trying to tune a disk performance issue on a NetBackup 7.1.04 system.

Media Server Deduplication (MSDP) is used.

Media Server Hardware: HP DL360 G7 2 x Quad-core CPU’s, 32TB RAM, Windows 2008 R2, SmartArray P812 (w/1024MB cache) controller for MSDP disk, QLogic Dual-port 8gbps fibre-channel HBA

MSDP Disk: 2 x 12 x 2TB Midline SAS drives in 2 x HP D2600 enclosures. Connected via SAS cables in HP recommended configuration (fully redundant, multiple pathways). RAID6

Tape: HP MSL4048 library with 2 x Ultrium 3000 LTO5 drives connected via 8gbps Fibre channel to f/c fabric.

The problem: Tape write performance is only reaching 25MB/s due to what appears to be a DISK READ bottleneck on the MSDP disk pool

I’ve collected perfmon stats over the weekend when all the tape backups are running and observed that:

MSDP DISK READ Queue is constantly above 3 (whereas DISK WRITE Queue is only 0.1)

MSDP % DISK READ time is 300 – 450% (whereas %DISK WRITE Time is less than 15%)

MSDP AVG DISK BYTES/READ = 90KB

MSDP AVG DISK BYTES/WRITE = 150 – 210KB

MEMORY AVAILABLE MEGABYTES = 9000 – 12000MB (9 – 12GB!)

Those stats suggest to me that the READ’s are the bottleneck, and there is a significant amount of RAM sitting unused.

So my questions are:

  1. Is there any NetBackup tuning recommendations to improve READ caching (either in Netbackup or in the Windows OS) to make better use of all that free RAM?
  2. SmartArray controller has an adjustable Read:Write ratio, currently set to default 25% Read : 75% Write. Would you recommend changing that to favour Reads eg. 50:50?
  3. What is the recommended RAID stripe size for a 32TB MSDP disk array (HP’s default stripe size is 256KB, but it can also be set to 16K, 32K, 64K, 128K)?

We have already tried increasing NBU Buffers:

NUMBER_DATA_BUFFERS = 512

SIZE_DATA_BUFFERS = 524288

NUMBER_DATA_BUFFERS_DISK = 512

SIZE_DATA_BUFFERS_DISK = 1048576

And we’ve reviewed the NBU 7.1 Tuning guide for Windows and these articles:

http://www.symantec.com/business/support/index?page=content&id=HOWTO61249&profileURL=https%3A%2F%2Fsymaccount-profile.symantec.com%2FSSO%2Findex.jsp%3FssoID%3D1351810548972izpWz6g0a4jOpPe4k7I0p0bI7dSjkQ33HdG6u

http://www.symantec.com/business/support/index?page=content&id=TECH145791

Any hints or tips would be appreciated.

Cheers

Comments 5 CommentsJump to latest comment

captain jack sparrow's picture

Did you tried with null test (bpbkar)?

also what is cr queue in MSDP

try to do ftp on those drives from either of your clients and check if similar performance is achieved or more/less

Not sure of NOSHM would be useful. but try if it helps

 Cheers !!!

CJS

warwick5's picture

Hi Captain Jack,

What is the best way of monitoring the CR queue? The duplications from MSDP to Tape run for over 48hrs over a weekend.

Haven’t tried bpbkar null test, but did use IOMeter when the system was first installed and the disk is capable of streaming Reads & Writes @ 1050MB/sec & 800MB/sec, and over 1100 IOPS (100% random, 67% READ)

captain jack sparrow's picture

MSDP to tape??? Is that rehydration process are you referring to ?

 Cheers !!!

CJS

captain jack sparrow's picture

7.5 is the option to move. Significant improvements have been seen in rehyderation method.

If upgrade is anywhere in timeline of your project / suggestions go for 7.5 with 0.4 patch

 Cheers !!!

CJS