HP StorageWorks Scalable File Share - Manual (Page 330)

Performance figures B–2
B.1 I/O performance
This section provides expected I/O performance figures for a single server in the HP SFS system. These
figures are based on tests carried out by HP.
Note the following points in regard to these figures:
• The
raw_lun_check.bash
script was used to obtain the raw performance figures. This script tests
the speed of reading and writing 4GB of data. The devices are unmounted and remounted between
the write and read tests.
• Figures similar to those shown for single-client Lustre performance can be obtained using the
ost_perf_check.bash
script. This script reads and writes 4GB of data for each OST device
tested. In the tests on which these figures are based, the tests were run on client nodes with less than
4GB of memory.
• Figures similar to those shown for multi-client Lustre performance can be obtained using the
IOR
sequential read and write benchmark on eight client nodes. Each client process reads or writes 2GB
of data to individual files in parallel.
• All results are based on tests over a Quadrics interconnect; however, given the raw speeds of Myrinet
2XP and Voltaire InfiniBand interconnects, there should not be any significant difference when these
interconnects are used.
• All results are based on tests using new, empty file systems.
EVA4000 storage
Table B-1 provides details of expected performance figures for one Object Storage Server in systems using
EVA4000 storage. These figures are based on the following configuration:
• Dual-connected QLogic host bus adapters (QLA2342 (from QLogic Corporation)—HP Part Number
FCA2214 DC HBA adapters).
• EVA4000 controller firmware version XCS v5.100.
• Two disk groups, each containing fourteen 146GB 10K RPM disks.
• One LUN per disk group.
In Table B-1, the first set of figures (1 LUN) for each entry provides the data for a scenario where the first
disk group is accessed; and the second set of figures (2 LUNs) provides data for a scenario where both disk
groups are simultaneously accessed.
Table B-1 Performance figures — EVA4000 storage
Number of LUNs Write (MB/sec) Read (MB/sec)
Raw Performance 1 170 185
2 280 350
Single-client Lustre
Performance
(one client,
ost_perf_check
command)
1 180 185
2 280 290
Multi-client Lustre
Performance
(eight clients,
IOR
utility)
1 180 185
2 280 350