MSA 1040 improve read speed?

[MarcPeterson]

All,

I've got an older iSCSI MSA 1040 that I'm currently using as a disk backup target in a disk-to-disk-to-tape scheme.  It's got (12) 7,200 rpm near-line drives in a One Big Raid 10 configuration, with a single volume, connected to an DL360 G9 w/2 onboard 1Gb per second NICs.  iSCSI MPIO is setup and configured correctly,  Jumbo frames enabled and verifed.  Its on the latest firmware, and setup with linear storage.

The tape drive is an LTO-8 unit hanging off an H241 SAS HBA (12Gb per second).  This drive has a minimum write speed of 100MB per second and a maximum of 300MB per second. 

The disk to disk portion of the backup is working as expected.  However, when spooling data off the MSA to the tape I'm seeing sustained transfer speeds (as reported in the backup software) of approx 66 MB per second.  This is causing the drive to stop, rewind, and start as the read buffers get empty.

Is there anything I can do to increase the speed here?

Thanks,

Marc Peterson

[JonPaul]

@MarcPeterson 
I think your bottleneck is the network port.  
With a single big RAID 10 on a 1040, you either have one Pool or one Linear disk-group.  The Pool/Disk-Group aligns with a single controller and MPIO is going to limit the server to only using a single array host port.  You should be able to see where your data is flowing through the CLI command:  show host-port-statistics
That means that your maximum throughput to the server is 1Gb (or ~100MB/sec).  If you are using this port for READ and WRITE at the  same time then 66MB/sec is probably pretty good.
Options for improving performance, which would mean removing what is likely the current bottleneck of the single 1Gb network port.
1. Increase the port from 1Gb to 10Gb  (If the 1040 was purchased as a 1Gb system then this is a non-starter)
2. Add another network port to the second 1040 host port (need both the server and the MSA to have another port)
3. Not recommended - intentionally defeat the MPIO so that the non-optimized port is active. This will improve throughput but at the expense of the controllers working harder to handle data i.e. lower performance out of each controller but more aggregated performance.
You should also do some investigation of the actual bottleneck CLI commands:
show controller-statistics
show host-port-statistics
show disk-statistics
Other statistics commands from the MSA could be helpful

[MarcPeterson · Accepted solution]

JonPaul,

I was able to login while a backup job was running and checked all the stats.  Nothing looked out of the ordinary, other than it was only doing about 40/50% of the capacity of each iSCSI link.

After the backup was completed I did a simple windows file transfer of a large file and even though the reported explorer window file transfer speed was inaccurate as a result of the caching, the network showed both links saturating at 1 gbps each.

I figured it was something I miscofigured so I went and did some tuning in the backup software, and I'm now seeing sustaned end-to-end througput of approx 188-210 MB/sec, which is very acceptable.

So it ended up not being related to the MSA.

Thanks for your time.

Marc