Poor performance - 2 DLT 7000 drives

[KenChew]

Hi,

I have a StorageTek TimberWolf 9730 with 2 DLT 7000 drives. I've configured parallelism to 4 and each device with a target session of 2. However, I notice that at each time, one of the drive would be performing at a very low rate of about 100-200 kb/s while the other is running at 5mb/s?
Can anyone please advise?

 

[rgzimmer]

Lots of things can affect tape performance. For example, I have a volume on a server that contaings 10gb of database files. One folder in that volume has over 500k text files in it, but is only around 1.5gb. It takes the server less time to back up the 8.5gb of large database files than it does the 1.5gb of numerous smaller files. While it writes those small files, it runs at around the same speeds you mention (150 - 200kb/s). When it finishes them, it goes back up to 10mb/sec (DLT 8000's)

You also may want to increase target sessions to 4. DLT's work better when they have something to do.

 

[renault5]

Hi KenChew , another cause of this problem (at my site atleast) is the problem of having the backup server connected to multiple networks (through multiple NICs) , in my case one network backs up fine (4-5Mb/s) but the other has to pass through a firewall which offers very restrictive
transfer rates - similar to your 150Kb/s.

Just an idea.

 

[wmg]

Ken,

We had a similar problem (although, we also were getting LOADS of read errors with the tapes). HP advised us, curiosly, that you shouldn't daisy-chain DLT7000 drives - each drive should have it's own SCSI bus (though one can share with the robot when you have a jukebox). Evidently, this worked fine with DLT4000 drives but, with DLT7000, the throughput is greater.

Since we have run separate SCSI cables to each of the drives and installed additional Host adaptors in the server, we have not had nearly as many issues with Read Errors, SCSI timeouts and the like.

Hope that helps.

Regards

 

[KenChew]

Hi,

I've increased the parallelism and the no. of sessions and it didn't help. I've also checked the NIC settings, all my servers are running 100mbs at full duplex and are on the same segment.
Although I'm encountering slow output, I'm not seeing Read Errors.
It seems that I'm getting better performance just using one drive at a time. But that's a waste of resource!
I desperately need to get more output as I'm expecting to put at least 10 more servers on the juke box.

 

[rgzimmer]

It sounds to me like the drives are fine, but the clients are the problem if the drives "appear" to run fine if used alone. The writing throughput that you see is cumulative for all streams to the drive. If you have one stream on one drive writing at 5mb/s and the other drive writing at 150kb/sec, it would appear that they write at 5150kb/sec if they both wrote to the same drive.

If possible, run a test against a server that has a large number of smaller files (in the hundreds of thousands). Do a full backup of the server, then run another backup with a directive to skip the volume/directory with the large number of files. I think you'll be amazed at the difference in write speeds.

Additionally, the client(s) may not be able to provide enough throughput to maximize drive utilization. My test backup server has two DLT 8000's. I have an old 1600 that I did some testing on with Novell backups. No matter how few or how many streams I sent out of that old 1600, I could never get it to send more than 3mb/s cumulatively, even writing across both drives. I got the same result when backing up to my primary Legato server from that client.

It may take quite a bit of research for you to get the optimum solution. Lord knows I've done a ton. Good luck.

 

[KenChew]

Hi all,
I finally solve the problem by changing the duplex setting on the switch and the NICs to half duplex. Thanks for all the advice given.