Multiple backup jobs at once

[KiwiJJ]

Hi,
Is Veeam able to do multiple backup jobs at once to write to a removeable disk drive ?

thanks,

John

[Gostev]

John, yes absolutely we can do that. However, for remote targets I do not recommend having too many concurrent jobs hitting the same target, as this results in target saturation issues, that in turn may lead to random backup failures due to I/O timeouts.

Also, based on my testing with local targets, unless we are talking about SSD drive on target, writing 2 files to the same storage in parallel often takes comparable or even more time than writing them sequentially (due to hard disk drive based storage specifics)... so there's no benefit... just something to keep in mind.

I've done such testing using Windows Explorer a few months ago, if you want you can try yourself. You need very fast source storage, preferably 3 similar disks: 2 source disks with the same file, and 1 target disk to write to. Try to copy them in parallel, then sequentially. Just don't forget to flush OS file system cache between tests, or use large files!

I actually bought SSD drives for my personal computers after this testing

[KiwiJJ]

Hi Gostev,
Thanks for the information I will give the testing a go and see how it works out..

regards,

John

[fgw]

have three jobs running at once. from looking at the read and write bandwidth you can see the target storge write bandwith is surprisingly low. during my backupsession the write bandwith peaks a few times for some seconds up to 130MB/s but stays most of the time far below 50MB/s. read bandwith on the other hand reaches about 300MB/s several time. at the same time there is just about 20MB/s write bandwith on the target storage!

and this picture is as i guess expected. taking into account, the backup server needs to read every single block to check if it needs to be backed up. even the blocks required to be backed up will be compressed before written to dist, thus the amount of data written is far below the amount of data read!

of coures, if you have multiple jobs running at once, you need raid disks to prevent race conditions where the job compete for access to the disk.

as a note: the bandwith figures here are not as reported by veeam backup itself. this figures are windows performance counters, thus reflecting real disk io. remeber veeam figures usually show average io: (total backup time / total vm filesize) which includes also times required for creating and removing snapshots, thus won's show actual disk load!

[Gostev]

and this picture is as i guess expected. taking into account, the backup server needs to read every single block to check if it needs to be backed up. even the blocks required to be backed up will be compressed before written to dist, thus the amount of data written is far below the amount of data read!

Yes, this is expected for VI3. However with vSphere, we do not need to read every single block, as ESX4 changed block tracking provides the information about all blocks changed since last pass instantly for us, so we only need to read and backup changed blocks.

Other than that yes, there will always be much fewer writes than reads, due to very significant data size reduction with compression and deduplication our product performs before putting the data into backup file.

[glennsantacruz]

As a side question - what software are you using to produce these graphs?

[fgw]

As a side question - what software are you using to produce these graphs?

RivaTuner together with a self written Plugin to read windows performance counters ...