5190/5194 Modular Tape Subsystem Manual

Operating the Cartridge Tape Drive and ACL
5190/5194 Modular Tape Subsystem Manual—426878-002
4-56
Using a 5190 or 5194 Drive for a Memory Dump
(NonStop Himalaya K-Series Servers)
applications must call CONTROL with the value 26 to flush all outstanding buffers to
tape. CONTROL returns an error if any buffers are not successfully written to tape.
Table 4-11 examines the performance trade-offs between buffering and error reporting.
Using a 5190 or 5194 Drive for a Memory Dump (NonStop
Himalaya K-Series Servers)
If you want to use a 5190 or 5194 tape drive to capture memory dumps, read the
following information.
NonStop Himalaya K2, K100, K200, K1000, K1000SE, K2000,
and K2000SE Server Memory Dumps
When you use the Copy Memory: Tape screen on the system control panel to dump
memory, the internal cartridge tape in the base system cabinet (cabinet 1) is used by
default. However, you can use a 5190 or 5194 tape drive for memory dumps (from the
control panel) if the system is configured in one of the following ways:
•
MFC connection
The 5190 or 5194 tape drive CRU is attached to an MFC that is connected to
processor 0 or processor 1 (the MFC must be in the base system cabinet or in I/O
cabinet 1A) and the tape drive is specified as the default drive to be used for
memory dumps. (You use TMDS MDS subsystem commands to do this. See the
TMDS Reference Manual.)
•
3216 controller connection
The 5190 or 5194 tape drive CRU is attached to a 3216 controller located in the
base system cabinet or in I/O cabinet 1A, and the drive is designated as unit 5 on
the 3216 controller (in the system configuration file). The tape drive must be
specified as the default drive to be used for memory dumps. (You use TMDS MDS
subsystem commands to do this. See the TMDS Reference Manual.)
Table 4-11. Buffered Mode Performance
Mode Performance
Buffering off
(record-level
recovery)
Performance is slowest for all file sizes. Error reporting, however, is
immediate; when a tape request fails, the error is associated with the
current record.
Buffer data
(file-level
recovery)
Performance is good for large files but poor for small files. Error reporting
is at the file level: when a buffered-tape request fails, the error is
associated with the current file and the application must recover the file.
Buffer data
and file marks
(reel-level
recovery)
Performance is best of the three modes. Error reporting is at the volume
level: when a buffered request fails, the error is associated with the current
volume or reel and the application must recover the entire volume.