Seagate ST3500641SV ユーザーズマニュアル
48
SV35 Series SATA Product Manual, Rev. C
If write caching is allowed, the actuator could be busy servicing a cached write at virtually any time. The impli-
cation is that a timeout on a read command could actually be due to a cached write. To address this case:
• Streaming commands will not time out if there is a non-streaming cached write in progress
• all cached write data is discarded if a timeout occurs on a streaming command
• when the host issues a streaming read to the drive, internal drive activity is limited to minimize the chances
cation is that a timeout on a read command could actually be due to a cached write. To address this case:
• Streaming commands will not time out if there is a non-streaming cached write in progress
• all cached write data is discarded if a timeout occurs on a streaming command
• when the host issues a streaming read to the drive, internal drive activity is limited to minimize the chances
that significant time will be spent on a normal write, and
• timeouts are inhibited if this case is occurring.
4.3.4.3.2
Non-cached streaming writes use model
The host can issue streaming writes. If this is done, it is recommended that the host either set the F bit or dis-
able write caching. If the host does neither, the drive has no way to accurately report the failure location and
length in the event of a write failure on a cached write. Disabling caching or using the F bit will cause some per-
formance loss compared to the case where caching was allowed. The most straightforward way to ensure that
performance is maintained in this usage model is for the host to issue large writes, so that the seek time is
insignificant when compared to the amount of time that the host can allow the command to consume. If the
host’s time limit allows, the drive will perform internal defect management on any defective sectors encoun-
tered. If the drive runs out of time during a write of this type, any currently cached sectors in the drive’s buffer
will be discarded. If the error occured in the LBA range of the current command (rather than a cached com-
mand), the portion of that write that may not have been successfully written to the media will be reported to the
host in the ending status of the write command.
able write caching. If the host does neither, the drive has no way to accurately report the failure location and
length in the event of a write failure on a cached write. Disabling caching or using the F bit will cause some per-
formance loss compared to the case where caching was allowed. The most straightforward way to ensure that
performance is maintained in this usage model is for the host to issue large writes, so that the seek time is
insignificant when compared to the amount of time that the host can allow the command to consume. If the
host’s time limit allows, the drive will perform internal defect management on any defective sectors encoun-
tered. If the drive runs out of time during a write of this type, any currently cached sectors in the drive’s buffer
will be discarded. If the error occured in the LBA range of the current command (rather than a cached com-
mand), the portion of that write that may not have been successfully written to the media will be reported to the
host in the ending status of the write command.
4.3.4.3.3
Non-streaming writes
In non-streaming writes, the drive does not discard any write data, but rather protects the data as if it were in a
non-timed environment altogether. The cache algorithm ensures that if a write to the media is begun, it will be
finished during the write command on the ATA bus, and will not adversely affect a streaming read.
non-timed environment altogether. The cache algorithm ensures that if a write to the media is begun, it will be
finished during the write command on the ATA bus, and will not adversely affect a streaming read.
A system may use this mode during boot, and then switch over to Streaming Writes for operational mode, back
again to untimed mode for maintenance, etc. This behavior can be safe as long as a Flush Cache command is
issued and completed before switching between write command types.
again to untimed mode for maintenance, etc. This behavior can be safe as long as a Flush Cache command is
issued and completed before switching between write command types.