On 12/5/23 10:24, Marco Gaiarin wrote:
I'm playing with my new LTO9 tape drive (Tandberg/IBM one, SAS). Some year ago, trying to find optimal parameters for an LTO-5 tape drive, i've found: options st buffer_kbs=16384 Minimum block size = 0 Maximum blocksize = 256K Maximum Network Buffer Size = 262144 Maximum File Size = 25G this lead to the LTO-9 unit: *speed file_size=16 nb_file=4 btape: btape.c:1062-0 Test with zero data, should give the maximum throughput. btape: btape.c:911-0 Begin writing 4 files of 17.17 GB with raw blocks of 262144 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 592.4 MB/s btape: btape.c:1074-0 Test with random data, should give the minimum throughput. btape: btape.c:911-0 Begin writing 4 files of 17.17 GB with raw blocks of 262144 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write
rate = 279.3 MB/s
btape: btape.c:1088-0 Test with zero data and bacula block structure. btape: btape.c:966-0 Begin writing 4 files of 17.17 GB with blocks of 262144 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 234.5 MB/s btape: btape.c:1100-0 Test with random data, should give the minimum throughput. btape: btape.c:966-0 Begin writing 4 files of 17.17 GB with blocks of 262144 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 223.8 MB/s Trying to double all the value, as a starting point: options st buffer_kbs=32768 Minimum block size = 0 Maximum blocksize = 512K Maximum Network Buffer Size = 524288 Maximum File Size = 50G lead to: *speed file_size=16 nb_file=4 btape: btape.c:1062-0 Test with zero data, should give the maximum throughput. btape: btape.c:911-0 Begin writing 4 files of 17.17 GB with
raw blocks of 524288 bytes.
[...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 660.7 MB/s btape: btape.c:1074-0 Test with random data, should give the minimum throughput. btape: btape.c:911-0 Begin writing 4 files of 17.17 GB with raw blocks of 524288 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 279.3 MB/s btape: btape.c:1088-0 Test with zero data and bacula block structure. btape: btape.c:966-0 Begin writing 4 files of 17.17 GB with blocks of 524288 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 247.1 MB/s btape: btape.c:1100-0 Test with random data, should give the minimum throughput. btape: btape.c:966-0 Begin writing 4 files of 17.17 GB with blocks of 524288 bytes. [...] btape: btape.c:385-0 Total Volume bytes=68.71 GB. Total Write rate = 237.7 MB/s So, a little better but not doubled the throughput
, and this clearly was
expected. So probably the optimal buffer size sit between 256K and 512K; but because server have plenty of RAM, i think my search will stop here. ;-) A queston rise on me: why 'bacula block structure' have a such great impact on hardware compression?! EG, why if i write zeroes in raw mode i get 660.7 MB/s while if i write zeroes in 'bacula block structure' i got 247.1 MB/s?! Compressione seems correctly enabled: root@svpve3:/etc/bacula# tapeinfo -f /dev/nst1 Product Type: Tape Drive Vendor ID: 'IBM ' Product ID: 'ULTRIUM-HH9 ' Revision: 'Q3F5' Attached Changer API: No MinBlock: 1 MaxBlock: 8388608 SCSI ID: 2 SCSI LUN: 0 Ready: yes BufferedMode: yes Medium Type: 0x98 Density Code: 0x60 BlockSize: 0 DataCompEnabled: yes DataCompCapable: yes DataDeCompEnabled: yes CompType: 0xff DeCompType: 0xff BOP: yes Block Position: 0 Partition 0 Remaining Kb
ytes: -1
Partition 0 Size in Kbytes: -1 ActivePartition: 0 EarlyWarningSize: 0 NumPartitions: 0 MaxPartitions: 3 Thanks.
Hello Marco, The following settings have been found to work quite well with the LTO9 drives I am testing in Quantum's lab: ----8<---- MaximumFileSize = 32GB MaximumBlockSize = 2097152 ----8<---- Also, do *not* set `Minimum block size`, and the `Maximum Network Buffer Size` is probably also not necessary. Hope this helps! Bill -- Bill Arlofski w...@protonmail.com
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
