Advanced Backup to Disk Performance White Paper

RAID 1 was only tested in the 1 logical array configuration, which resulted in poor performance
values (write 9,24 MB/s and read 15,72 MB/s). Furthermore, it must be considered that RAID 1
reduces the usable disk space by 50%.
RAID 1+0 achieved better results than RAID 1 (write 16,96 MB/s and read 29,26 & 48,30 MB/s)
due to striped disk data but the problem of inefficient disk usage remains.
RAID 5 was only tested in the configuration with 2 logical arrays to follow the disk array technology
with 2 SCSI/SATA ASICs. 6 physical disks were grouped into one logical array. If writing to 1 LUN,
the transfer rate showed 21,33 MB/s. If writing simultaneously to 2 LUNs, the transfer rate increased
to 34,43 MB/s. If reading from a single LUN, the read transfer rate was 36,57 MB/s. If
simultaneously reading from 2 LUNs, the read transfer rate was 56,64 MB/s.
The RAID 5 level is chosen for further testing. The results give a good estimate what the MSA1500
staging area (file library) is able to deliver. The maximum read performance of 56,64 MB/s is a
good prerequisite for utilizing two SDLT 320 (2*35 MB/s, 2:1 comp.) and Ultrium 460 (60 MB/s,
2:1 comp.). But it shows similar to the MSA1000, that this best-case scenario cannot completely
utilize the high-performance Ultrium 960 (157 MB/s, 2:1 comp.).
Backup to NULL Device
The backup to the NULL device is a good proof point for the maximum read performance and which
tape drive technology would fit to it.
The best typical file backup was achieved with the concurrency of 4. The transfer rate of 92,80 MB/s
matches almost the MSA1000 read test result of 93,58 MB/s. The results for small files are similar.
Please check
Table6 and Table8.
The initial backup causes Data Protector to save file information into the internal database, which
requires more system resources. The initial backup of 22 million small files showed that even such a
high number of files have only little impact on the CPU load (27% instead of 23%). Please check
Table8.
Backup to Tape
The SDLT 320 saved the data from a single LUN to tape with 28,84 MB/s, the Ultrium 460 with
39,39 MB/s and the Ultrium 960 with 43,01 MB/s. Both Ultrium (460/960) were slowed down due
to the limited disk read performance of 45,70 MB/s. Please check
Table6, Table9, Table10 and
Table11.
The Ultrium 960 was tested with different block sizes (64 and 256 KB). The data transfer rate was
comparable (91,15 and 91,97 MB/s) but the average CPU load was much different (23% and 11%).
Finally, the increased block size of 256 KB reduced the CPU utilization by 50%. Please check
Table11 and the figures below.
37