10-16-2014 03:01 AM
HI there,
We are in the process to make some architecture changes to the current NBU environment, simply put there will be a media server that has existed that will be taking the brunt of the jobs churning and policies running instead of a balanced load, I have read some documentation on Performance tuning and buffer sizes and that for disk vs tape etc...
I wanted another fresh perspective from some of you that have been able to produce some pretty good/significant tuning and a level that is acceptable and churns at optimal speed. Now I am aware that there are various parameters to set and I wanted to check if somebody has a new white paper on this or some general tips, please don't post a link to the official admin guide for NBU that is not the kind of input I want from the people on here, I want their experience.
Some pre-req info to assist in this.
SERVER 2008 R2 running 7.6.0.2 across all master/media's
separate backup lan able to do 1Gbps connections, media server has x4 1gbps nics teamed in aggregated 4gbps tunnel
ML6000 library with LT04 FC connected drives to Brocade
In your experience with these technologies, what is your disk_buffer and buffer_x sizes and what have you found works well to tune on the media server
its a combination of tape out and disk_staging happening on this media and as I said it will be taking a much larger hit soon and churning a fair bit
All the clients connecting in on the backup lan has teaming set on the respective clients running at 1Gbps as well and a separation between BACKUP LAN and PRODUCTION LAN - host.ini files edited to force backup traffic over separate subnet.
Again im looking for some personal opinions and "real life" examples of what has and has not worked for some of you.
10-18-2014 12:28 AM
Be aware of the way the switch allocates the links the tunnel, have seen 4Gbps trunks that never came over 1 Gbps because the allocations was based on the ip of the recieving media server
For LTO4 drives I would say at least a block size of 256KB (262144) and the recommendation for disk is 1 MB (1048576), the numbers of buffers depends on the infrastructure but would start with at least 64 then increase untill the waiting messages has almost disappeared. We ended up stop tuning our LTO5 tape drives as our old disk system couldn't handle the read load from the backups. But with 256 KB & 256 buffers we had a pretty good throughput overall.
Also look at the technote for tuning windows for netbackup, think it was originally for NBU 7.0, but some of the OS stuff is still valid
In general found that you need to tune the underlying hardware & OS layers, specially the network for a good NetBackup performance. Have used Iperf for windows to get an idea of what throughput I could get with multiple streams from multiple clients
Another thing I have seen is that the image db disk reponse times have a big impact as netbackup else have to wait on the catalog to get updated. Giving the NBDB it's own disk can improve netbackup media/resource handling
There is also some stuff that can be done in nbrb.conf in relation to resource allocation, but think that requires a support to be supported
11-06-2014 09:42 PM
Performance Tuning best practice session at Vision earlier this year: