Availability Guide for Application Design

Overview of Server and Network Fault Tolerance
Availability Guide for Application Design—525637-004
2-14
System Process Pairs
Checkpointing Data in the Disk Process
The need to checkpoint data depends on whether the operation is retryable. For a
retryable operation, such as a read or a full sector write, there is no need to checkpoint
any data. The file system can simply retry the operation on behalf of the application.
The extent to which partial sector writes are retryable depends on whether the file is
audited. TMF software provides audit protection for operations that take place in
transaction mode on audited files. Audit protection keeps a copy of the data record or
row before the file was updated.
In the case of audited files, if the primary disk process fails, the backup disk process
receives the resent request from the file system. The backup disk process can apply
the request to the before-image data before writing the record or row to disk. In the
case of audited files, the only information that need be checkpointed to the backup disk
process is a transaction identifier and record lock information. For more information on
audited files, refer to Section 4, Data Protection and Recovery.
For nonretryable operations, such as a partial sector write to a nonaudited file, the disk
process must checkpoint the entire updated segment before writing to the disk. Once
the checkpoint is acknowledged, the primary process can go ahead and write out the
data; the backup process has a full sector available for retry.
Figure 2-5. A Disk Process: An Example of an I/O Process Pair
Processor 0 Processor 1 Processor 2
ServerNet
Adapter
Dual ServerNet Fabrics
Mirrored Disks
Device
tables
Application
process
Primary disk
process
Backup disk
process
File system
VST205.vdd
ServerNet
Adapter