ProLiant Servers - Netservers

Abysmal read performance from ML350 G5

 
Jeremy Shetland
New Member

Abysmal read performance from ML350 G5

It's using the Smart Array E200i controller, with the 128MB BBWC upgrade. Xeon 5140 (2.33), 4 GB, 5 HDD's, Datacenter Server x64 Ed.

It could be the onboard E200i is too weak for what I'm running. First 3 drives are RAID5'd, with two partitions (OS and user data), another drive has 2 VM's, and a fifth drive has another 2 VM's. That sounds really heavy, but the VM's are small, and about 15 users are currently using the file server. Memory use peaks at around 2.5 GB. ("Available" physical memory is around 600 MB). This problem came to my attention when I configured a backup which took 33 hours to complete - older servers backed up at around 168 MB/min, while the new server averaged 13.8 MB/min. (Backup Exec agent is in use on all servers and media server has a 100Mb/full connection to the same fabric the other servers are on.)

Here's a test I ran. I copied a 176 MB folder containing a mix of small and medium-size files. First, from an older server to my PC, over a 100Mb/s port. It took 40 seconds. I copied the same folder from my PC to the new server, it took 25 seconds. I then copied the same folder from the new server back to my PC - 3 minutes and 35 seconds.

Where should I start looking for problems? I can't spend money on more RAM or another controller unless I'm sure it's going to fix the problem. Cheers!
1 REPLY 1
Jeremy Shetland
New Member

Re: Abysmal read performance from ML350 G5

As a side note - it makes no difference if the share I'm copying to and from resides on the raid array or on a lone disk.

And another fun fact -
Folders will copy between partitions at expected rates - a 500 MB folder will copy to the RAID 5 array in about 30 seconds, and will copy to a lone disk in 15 seconds. This is a relief, since it means the SAS controller isn't too much a bottleneck - I think?

Free points and my undying love to anyone who answers :p Come on - I bet it's something obvious.