Poor Write Performance with MSA2312fc as B2D Device

[Martin Widemann]

Hello,

i´ve done some Troubleshooting with a MSA2312fc Storage Array at a customers site, where we figured out a very bad write Performance with Backup Exec Backup to Disk. Here some Informations about the Configuration:

MSA2312fc with two vDisks, one as Raid5 with 16 Disks and one with Raid5 with 6 Disks. On the vDisk with 16 Disks is a Volume with 14TB mapped over Controller A to one Host, on the seconde vDisk is a Volume with ca. 5TB mapped to the same Host over Controller B. Owner of the vDisk is allways the Same Controller as the Controller for the Volume. Stripe Size is 64KB.

When we now start a Backup to Disk on the Big Volume we have the Poor Performance of Round about 30MB/s but on the small Volume Round about 4 Times More.

Did somebody know where the Bottleneck could reside? Is the Problem the vDisk with 16 Disks or the Size of the Volume?

Normally i work with EMC Storage and on EMC Storage i never whould design a Raid Group with 16 Disks from a Performance point of view, but i don´t know how the MSA Array internally works and the Informations i found on the WWW are not very helpfull

[Eemans Dany]

Hi,

Check if the Read/write cache is enabled on the disk/lun where you have problems. Otherwise i would sugest to create not a lun with 16 disks but 2 luns of 8 disks.

Dany

[Martin Widemann]

The Cache is enabled (write-back) with Super Sequential Mode. Thats the Information i found in the Best practise PDF for the MSA2000 G2.

[Fran Garcia]

Am I the only one that finds a 16 disks raid just horrid? Specially being SATA disks!

[DataCenterOperator]

we have raid 50 with 30x 1TB sata disks 8tb luns, we use netbackup b2d under sles11 64bit and have over 100MB/s. msa2312fc dual controller and proliant dl380g5 make round robin over 2hba(4ports).
read default
write super sequentiell

read best practise:
http://www.hpcollateral.com/Files/EcoCollateral_20090420_Whitepaper_MSA_Best_Practice__ipcz.pdf

[Martin Widemann]

What is your Chunk Size? Does Netbackup also writes with 32KB as Backup Exec does?

After Changing the Disk Config from 16 Disks in one vDisk to 2 vDisks with 8 each we figured out a Performance Increase. But we only have a write Performance on a 800GB Job of 21GB per hour...........

The Cache Settings i will try again, but with the 16 Disks vDisk there are no signifikant Changes from the Cache Settings.

[DataCenterOperator]

netbackup chunksize is 256k, but better it's same as vdisk or half on it.
default is 64k on one disk, if you have 16disks in raid 5 then 64k*16-1 = 960k