performance issue --- transfer rate dropped down not cames back

[redfox]

Hi all,

I recently got involved in a backup requirement for 20TB data from one unix client in 40 hours, I tried to split each job around 30G or 80G, and I got handreds jobs to run. my network band width is about 2 GB and I don't know the disk transfer rate. if I start 10 streams at the same time in media server A, each stream got 10MB/s and when I tried to submit more jobs utilizing other media servers, I noticed the early streams in server A dropped down to 7M/s or 5 M/s and looks it depends on how many streams I submitted in other media servers, but not matter what I submitted the total speed remains 100MB/sec, so I just guess our disk transfer rate hit bottleneck about 100M/s.

the strange thing is when jobs ran in other media servers finished, the job I strarted first in media server A almost dropped to 2MB/s and it never comes back to 10 MB/s, apparently we have at least 100MB/s like the begining, even after 9 jobs finished in media server A the remain job still keeps running under 2 MB/s, why did it act this way, do we have any command to speed up the remain jobs?

thanks!

Fox 

[ahlip]

Hi,

Typically, the performance of the throughput is dynamic, the client does get restricted to a lower throughput due to Media server throughput bottleneck, but it should use the higher available bandwidth once the other jobs are finished.

The slow backup could be backing up large files in the earlier stage (giving to 7MB/s speed) then near the end when you saw slow backups, it could now be backing up small files (dramatically slower).

[Andy_Welburn]

Also, if you are monitoring this thru'put via the NB GUI, it will show an average Kb/s as opposed to current. So, if it has been sometime running at 2Mb/s it may take a while for the thru'put rate to apparently increase.

If you're actually monitoring current thru'put using some other method then I'm not sure what the issue could be apart from what 'ahlip' said in the previous response.

[redfox]

hi guys, thanks for your response. yes I am monitoring from detailed status of NB GUI, however, in my case I don't have a lot of small files, what I have are only 8 10G raw device files (db archive log of HP client) for a 80G job. by the way, I am using NBU 5.1mp5, I am wondering if it is a known issue for this version.