Disk I/O Throughput - IBM 170 Servers Manual

A storage space which is linked as shared, or a disk with caching disabled, requires more CPU to
process write operations (approx. 45%).
Sequential operations cost approximately 10% less than the random I/O results shown above.
Even though a Windows disk driver may have write cache enabled, some Windows applications may
request to bypass the cache for some operations (extended writes), and these operations would incur
the higher CPW cost.
If your application load is skewed to the upper or lower average disk operations, you may encounter a
smaller or larger CPW cost than indicated by the "rules of thumb".

17.5 Disk I/O Throughput

The chart below compares the throughput performance characteristics of the current IXA product against
the new iSCSI solution. The charts indicates an approximate capacity of a single target HBA when
running various sizes and types of random operations.
The chart below also demonstrates that the new page based architecture utilized in the iSCSI solution
provides better over all performance in Write and Read scenarios. While the sector based architecture of
the current IXA provides slightly better performance in small block Write operations, that difference is
quickly reversed when block sizes reach 4k. 4k block sizes and higher are representative of more I/O
server scenarios.
As with all performance analysis, the actual values that you will achieve are dependent on a number of
variables including workload, network traffic, etc.
160
140
120
100
80
60
40
20
0
IBM i 6.1 Performance Capabilities Reference - January/April/October 2008
©
Copyright IBM Corp. 2008
.
iSCSI Target, IXA Capacity Comparison
MB per Second
IXA
iSCSI 1 Targ 4 Init Std
Chapter 17 - Integrated BladeCenter and System x Performance
iSCSI 1 Targ 1 Init Std
iSCSI 1 Targ 4 Init Jumbo
288