After the first backup (a differential) we found the backup server CPU (max 85%) and RAM (around 60%) handled having 10 simultaneous streams. The 1GB NIC appeared to be 50% utilised. What was interesting was that the MSA1500 being used for disk staging was hammered. Current Disk Queue Length was spiking quite frequently at 27 … well above the recommended maximum of 2! Seeing as the system isn’t interactive we’re not too worried but the disk (RAID 5 … slow, I know) or controller (shared with other systems that are heavily used at night) could be our bottlenecks. This new bottleneck was not considered in our estimations. I never though disk would be a bottleneck before the NIC. A comparison against a similar job saw agent-tape being 8.5 hours (total 8.5), agent-disk being 3.5 hours and disk to tape being 1 hour (total 4.5 hours). That’s nearly a 50% reduction on our impact on production systems for backup. We want to find our optimum setting for concurrent streams. The only true test is to suck it and see. Tonight, we will run 15 streams at once. We’ll then compare against a previous Tuesday differential job from another week and see what the time savings were. After that, we’ll try 5 streams and see what the trends are.
I mentioned that we had to call in tapes for file recovery. The file server is Windows 2003 and I would normally utilise Volume Shadow Copy and make use fo the Previous Versions Client for operational file recoveries. However, the file server is struggling with disk space so this is not an option.