Hello,
We get quite low write rates on LTO9 drives during the current tape challenge: lower than 200MB/s when both drives controlled by the CTA tape server are mounted.
LTO9 drive perf just doing a “dd” to tape is 400MB/sec as expected. Why the performance is slow with EOS we dont understand. Is there something obvious we are missing in the tapeserver config?
Thanks,
George
Hi George,
as I just replied to Tim:
The write performance of our LTO9 drives is around 350 MB/s.The read performace varies a lot more but is usually around 200-250 MB/s.
We didn’t enable anything specific for those drives, we are using the same configuration as for IBM TS1160 tape drives.
Cheers,
Vladimir Bahyl
Hi George,
You say that the transfer rate drops by 50% when both drives are mounted—but you get full speed when only one drive is mounted?
This could suggest that the problem is contention on the disk system rather than anything in the tape server configuration. In your presentation you said the disks were all SSDs so I would not have expected any problems. However, if you want to isolate the problem to tape or disk, you could try simultaneously recalling from two tapes to different disk systems and measure what transfer rates you get.
At CERN we have one tape drive = one physical tape server = one tape server daemon. We have never tried deploying with the configuration you are using, so we are interested to hear what you learn.
Cheers,
Michael
Hi Vlado and Michael,
Many thanks for your replies. We actually found out that the network performance between the EOS nodes and the LTO tape servers was
much lower compared with TS1160 tape servers. After swapping network cables, the network perf was much improved and now we get reading rates around 220-230MB/s
George