FWeston
asked on
Expected throughput for LTO4 tape?
We have an Apple XServe RAID with 14 x 750GB SATA drives connected via fibre-channel to a Qlogic 5602Q switch. The XServe RAID is configured as two 4.1 TB volumes. Also connected to the switch is a Mac running 10.4. This Mac has a 2 port 4Gbps fibre-channel HBA with port 1 connected to the switch and port 0 directly connected to a Tandberg 224 LTO4 library. The Mac runs Retrospect 6.1 workgroup edition to backup the two 4.1 TB volumes on the XServe RAID to tape.
This works, but it seems like the backups are taking longer than they should. When a job runs, it indicates a throughput of approx 3200 MB/min. Wikipedia lists the max throughput of LTO4 as 120 MB/s (7200 MB/min), which is over double the speed I'm seeing. I normally expect real world numbers to be 15-20% lower than maximum values, but this is 50+ percent slower than the spec.
For comparison, I looked at the logs for another server which is running Backup Exec with an LTO3 drive and it backs up at 4200+ MB/min. I assume there's a problem somewhere and these are not expected numbers?
This works, but it seems like the backups are taking longer than they should. When a job runs, it indicates a throughput of approx 3200 MB/min. Wikipedia lists the max throughput of LTO4 as 120 MB/s (7200 MB/min), which is over double the speed I'm seeing. I normally expect real world numbers to be 15-20% lower than maximum values, but this is 50+ percent slower than the spec.
For comparison, I looked at the logs for another server which is running Backup Exec with an LTO3 drive and it backs up at 4200+ MB/min. I assume there's a problem somewhere and these are not expected numbers?
Britt Thompson
This may be a limitation of the speed of the drives in your XServe RAID...7200RPM drives would only be expected to have a throughput of around 105 MB/s based on my experience. Also, through experience, the XServe RAIDs have an overall slower performance in general.
ASKER
Still - 105MB/s should still yield much more than the 3200MB/min I'm seeing.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
No, the Backup Exec is also on fibre-channel but it's backing up from an HP EVA which probably has better throughput than the XServe RAID. We had to stop the backup this morning so people could work, and the average transfer speed was about 2.2 GB/min, which I'm gathering is about average?
2.2GB/min is very fast for 7200RPM dirves...much much faster than my current setup which may be related the speed of the machine that's running the backup.
One more thing, the LTO4 is still relatively new and the drivers may improve over time or may be much better with the new Retrospect 8 that's just come out.
One more thing, the LTO4 is still relatively new and the drivers may improve over time or may be much better with the new Retrospect 8 that's just come out.
SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
I'm very surprised at experts stating that 3200mb/sec and 2200mb/sec are "very fast". They are acceptable, yes, but no way would i label them "very fast," the latter especially is almost USB speed, which is a crawl. Modern 7200rpm drives can push out 100+mb/sec without breaking a sweat. Since the OP's drives are also in a RAID, the performance of these drives should not be in question whatsoever. (I have a RAID 5 utilizing 4 so-called Green drives, that run 5400-5900RPM. I can get ~300mb/sec reads)
Although I can't help the OP directly, I can provide reasurrance that faster speeds are possible and I have first-hand experienced them. I get ~5500mb/min writing to LTO4 from a single off-shelf SATA 1.5TB drive. I am using 2003 Server, Backup Exec 12.5, a Quantum LTO4 HH SAS drive and an LSI SAS HBA. The performance is definitely possible.
I have once had slow performance and it turned out my data was heavily fragmented. HFS volumes can be fragmented just as easily as NTFS. I would try defragging the RAID. Further, you could copy some of its contents to a fresh drive, which essentially will be defragged since it will be written sequentially. Then write to tape from the fresh drive and see if there's a performance increase.
Although I can't help the OP directly, I can provide reasurrance that faster speeds are possible and I have first-hand experienced them. I get ~5500mb/min writing to LTO4 from a single off-shelf SATA 1.5TB drive. I am using 2003 Server, Backup Exec 12.5, a Quantum LTO4 HH SAS drive and an LSI SAS HBA. The performance is definitely possible.
I have once had slow performance and it turned out my data was heavily fragmented. HFS volumes can be fragmented just as easily as NTFS. I would try defragging the RAID. Further, you could copy some of its contents to a fresh drive, which essentially will be defragged since it will be written sequentially. Then write to tape from the fresh drive and see if there's a performance increase.
ASKER
For what it's worth, we ended up sending the tandberg drive back to apple because we had nothing but problems with it. We went through 3 or 4 of them in as many months, tandberg support was stumped, and we eventually just got tired of not having a working backup strategy so we bought a promise vTrak array and are doing disk to disk backups now. The library would work for a while, then it would just "disappear" on the host, couldn't see it in retrospect or system profiler. The only way to fix it at that point was to cold boot the library, and then cold boot the Mac. Don't know if it was something to do with the library or the Mac, but the flakyness could have had something to do with the "low" performance.