 |
|
The measurements compare LVM2 and EVMS logical volumes both
using single path, two paths with a failover profile (a primary
path, and one fail over path), or multibus profile (using
two paths in a round-robin manner).
The measurements emphasize on throughput and CPU consumption
per KiB/s.
We used 16 SCSI disks, one disk from each rank.
|
|
|
- Initial write
-
| Single path: |
Both LVM2 and EVMS created LVs have the same
throughput ranges. |
| Failover: |
LVM2 created LVs show better throughput rates
than EVMS created LVs. |
| Multibus: |
Both LVM2 and EVMS created LVs have the same
throughput ranges. |
-
- Rewrite
-
| Single path: |
Both LVM2 and EVMS created LVs have the same
throughput ranges. |
| Failover: |
LVM2 created LVs show up to 20% higher better
throughput rates than EVMS created LVs. |
| Multibus: |
Both LVM2 and EVMS created LVs have the same
throughput ranges. |
-
- Initial write and rewrite costs
-
In failover and multibus mode, EVMS created LVs
show significantly higher costs than LVM2 created
LVs.
In single path mode, EVMS performs well with 1 process;
with more than 1 process the difference in costs
is less significant.
The costs are measured relative to single path LVM2.
|
|
- Read
-
| Single path: |
Both LVM2 and EVMS created LVs have almost
the same throughput ranges.
Exception: with 4 processes LVM2 created LVs
are up to 10% better. |
| Failover: |
Both LVM2 and EVMS created LVs
have almost the same throughput ranges.
Exception: with 4 processes LVM2 created LVs
are up to 10% better. |
| Multibus: |
Both LVM2 and EVMS created LVs have almost
the same throughput ranges. |
-
- Read costs
-
In failover and multibus mode, EVMS created LVs
show significantly higher costs than LVM2 created
LVs.
In single path mode, there is no big difference
between LVM2 and EVMS.
The costs are measured relative to single path LVM2.
|
|
- Random write
-
| Single path: |
The differences of LVM2 and EVMS created LVs
are not remarkable, they both stay in an acceptable
range. |
| Failover: |
For 4 and 8 processes, LVM2 created LVs perform
more than 10% better than EVMS created LVs. |
| Multibus: |
Both LVM2 and EVMS created LVs have almost
the same throughput ranges. |
-
- Random read
-
| Single path: |
Both LVM2 and EVMS created LVs have almost
the same throughput ranges. |
| Failover: |
The measurements are similar to single path,
with a bit lower throughput. |
| Multibus: |
For 1 to 4 processes, the measurements are
similar to single path and failover, the throughput
has the same range.
For 8 to 64 processes, the measurements degrade
and stay 30% below single path and failover
throughput rates. |
-
- Random read and random write costs
-
In failover and multibus mode, EVMS created LVs
show significantly higher costs than LVM2 created
LVs.
In single path mode, there is no big difference
between LVM2 and EVMS.
The costs are measured relative to single path LVM2.
|
|
|
- System z:
- 2084-B16 (z990) LPAR
0.83ns (1.20GHz)
2 Books each with 8 CPUs
2 * 32 MB L2 Cache
80 GB Memory
8 FCP Adapter (2Gbps)
- DS8300:
- 2107-922 (DS8300)
256 GB Cache // 1-8 GB NVS
256 * 73 GB disks 15.000 RPM
capacity: 23.3 TB (11.6TB SCSI, 11.6TB ECKD)
16 disk enclosures - 4 device adapter pairs
16 ranks SCSI disks
8 FCP (2 Gbps)
µCode Level 6.1.0.44
(RAID 5)
- Linux:
- SUSE SLES9 SP2
Linux 2.6.5-7.191-s390x #1 SMP Tue Jun
28 14:58:56 UTC 2005 s390x GNU/Linux
8 CPUs
8 FCP
256 MB
- Workload:
- Iozone 3.96 (http://www.iozone.org)
-
-
|
|
|
|
|
