IBM High Performance Storage System HPSS, Руководство по установке, Страница: 111 / 311

guaranteed that the file is completely and successfully staged before it is read. If the stage operation
fails, the open will return with an error.
Guideline 3: Select the Stage on Open Async option if you wish to stage the entire file to the top
level in the hierarchy and do not want the Client API open to block.
Explanation : When this option is selected, the file is staged in sections and the read and write
operations that access this file are blocked only until the portion of the file they reference has been
completely staged. Normally, this option is selected when the top level of the hierarchy is disk and
the files in this COS are fairly large in size. This option is only available when the top level in the
hierarchy is disk. If the top level is tape and this option is specified, the
No Stage option will be used
instead.
Guideline 4 : Select the Stage on Open Background option if you want the stage operation to be
queued in the Core Server and processed by a background Core Server thread.
Explanation : The Client API open request will return with success if the file is already staged.
Otherwise, a stage request is placed in a queue and will be processed by the Core Server in the
background. A busy error is returned to the caller. This option allows a large number of stages (up to
2000) to be queued in the Core Server and processed as resources permit. The other stage options will
result with a busy error if Core Server threads are not immediately available to process the request.
Guideline 5 : Select the Retry Stage Failures from Secondary Copy option if you want to configure
a Storage Class so that stage failures of the primary copy can be automatically retried from a valid
second copy. The associated hierarchy must first be configured for multiple copies.
Explanation : When a stage from the primary copy fails and a secondary copy of the file is available,
HPSS will usually reissue the stage operation from the secondary copy. This is typically done when a
tape, holding the primary copy, becomes damaged and cannot be read. A warning that the stage
operation has used the secondary copy will appear in the SSM Alarms and Events window.
3.10.3.4. Selecting Optimum Access Size
This field is only advisory in nature; however, for later releases it may be used by interfaces to
dynamically select good buffer sizes.
Guideline 1 : Generally, if the file is being staged on open, Optimum Access Size should be set to
the same value that Optimum Access Size is set to in the Storage Class at the top of the hierarchy.
Guideline 2 : If data is not being staged to the top level before it is read (either automatically or by
user command), select a value that is an integer multiple of the largest Optimum Access Size field
found among the storage classes that make up this hierarchy.
3.10.3.5. Selecting Average Latency
This field can be used via the COS Hints mechanism to affect COS selection.
Guideline 1 : This field should generally be set to the value of the Average Latency field in the
Storage Class at the top level of the hierarchy when the Stage on Open option is in effect. If files are
usually accessed only once after they are staged, the average latency should be set to the latency of
the level they are staged from.
Guideline 2 : If it is expected that most of the requests for files in this COS are read requests, then it
may be best to set the value of this field equal to the Average Latency field in the Storage Class in
the hierarchy where most of the data accesses come from.
Guideline 3 : If files are written into the COS much more frequently than read, use the Average
HPSS Installation Guide July 2008
Release 6.2 (Revision 2.0) 111