IBM C2B 2.25 Brick On Sled carrier 128-pin HPC - Manual (Page 44)

USER RESPONSIBLE F O R V E R I F Y I N G VERSION A N D COMPLETENESS
O E M F U N C T I O N A L SPECIFICATION ULTRASTAR X P (DFHC) SSA M O D E L S 1.12/2.25 GB - 1.0" H I G H
3.4.1 For Different Transfer Sizes
The primary performance change due to a change of transfer size is the Data Transfer to/from Disk param-
eter. See 3.3.1, “Basic Component Descriptions” on page 40 for an explanation of the calculation of this
parameter.
The Command Execution Overhead may also change if the transfer size is reduced to the point where certain
internal control functions can no longer be overlapped with either the SSA Link or Disk data transfer.
For example, a short read may incur up to .65ms extra overhead if the Data Ready/Reply exchange does not
overlap the disk transfer.
3.4.2 When Read Caching is Enabled
For read commands with Read Caching Enabled Command Execution time can be approximated by
deleting Seek, Latency, and Data Transfer to/from Disk components if all of the requested data is available
in a cache segment (cache hit). Command Execution Overhead increases by approximately .1ms in this case
as there is no overlap with seek/latency.
When some, but not all, of the requested data is available in a cache segment (partial cache hit) Data
Transfer to/from Disk will be reduced but not eliminated.
Seek and Latency may or may not be reduced
depending upon the location of requested data not in the cache and location of the read/write heads at the
time the command was received.
The contribution of the Data Transfer to/from SSA link to the Command Execution time may increase since
a larger, or entire, portion of the transfer may no longer be overlapped with the components that were
reduced.
3.4.3 When Write Caching is Enabled
For write commands with the Write Caching Enabled (WCE) Mode parameter bit set, Command Execution
time can be approximated by deleting Seek, Latency, and Data Transfer to/from Disk components.
The
contribution of the Data Transfer to/from SSA link to the Command Execution time may increase since a
larger, or entire, portion of the transfer may no longer be overlapped with the components that were
reduced. The reduced times effectively are added to the Post Command Processing Time.
Command completion status is returned when data is completely stored in the buffer. The time to transfer
this group of data to the disk will be added to the performance of any next command that was in the queue.
3.4.4 When Adaptive Caching is Enabled
The Adaptive Caching feature attempts to increase Read Cache hit ratios by monitoring workload and
adjusting cache control parameters, normally determined by the using system via the Mode Parameters, with
algorithms using the collected workload information.
3.4.5 When Read-ahead is Enabled
If read-ahead is active, the service time is affected in several ways:
If the data requested by a read command is all in the data buffer already, the command can be serviced
very quickly.
If the beginning of the requested data is in the buffer, and the read-ahead is still in progress, data transfer
Source filename=PERFORM IBM Corporation Page 44 of 87