Large Volume CMS Minidisks
In z/VM 4.3.0 the performance of the CMS file system was enhanced to reduce the amount of DASD I/O performed at file commit time. Previously, the entire file allocation map was rewritten to DASD every time a file was committed. The file allocation map is one of the control files residing on every CMS minidisk. The map contains one bit for every minidisk block indicating the allocation status of the block. In z/VM 4.3.0 the code was changed to write only the modified allocation blocks when the file is committed.
This section presents and discusses measurements that illustrate the performance effects of this improvement.
The CMS file system commands CREATE, COPY, RENAME and ERASE, all of which commit the file at end of command, were used to validate the performance improvement. All four commands were each issued fifty times in both z/VM 4.2.0 and z/VM 4.3.0 on files composed of 10 blocks. The use of 10-block files allowed the scenarios to remain the same for the different size minidisks. The number of DASD I/Os was measured using the number of virtual I/Os displayed by the CP QUERY INDICATE USER (EXP command. The measurements were recorded using different size minidisks formatted with a block size of 4096. Minidisks formatted with a smaller block size will produce similar results.
The measurement results are summarized in Table 1.
Table 1. Large Volume Minidisk DASD I/O Reduction
z/VM Level | 4.2.0 | 4.3.0 |
|---|---|---|
10 Cylinders CREATE COPY RENAME ERASE |
|
|
100 Cylinders CREATE COPY RENAME ERASE |
|
|
1000 Cylinders CREATE COPY RENAME ERASE |
|
|
3338 Cylinders CREATE COPY RENAME ERASE |
|
|
10020 Cylinders CREATE COPY RENAME ERASE |
|
|
32760 Cylinders CREATE COPY RENAME ERASE |
|
|
| Note: For 10-3338 Cylinders, 3390 Mod 3 DASD used. For 10020-32760 Cylinders, Shark Large Volume DASD used. All minidisks were formatted with a block size of 4096. | ||
In the base case, z/VM 4.2.0, the number of virtual I/Os increases as the size of the minidisk increases. Any change to the file allocation map causes the entire map to be rewritten to DASD. This is shown in Table 1 by the increase in the number of virtual I/Os as the size of the minidisk gets larger.
With the new performance enhancement in z/VM 4.3.0, the number of virtual I/Os remains almost constant as the size of the minidisk gets larger. In this case, writing only the modified file allocation blocks has decreased the number of DASD I/Os necessary for the large cylinder minidisks.
As seen in Table 1, the benefits of the CMS Large Volume Minidisk improvement does not impact small minidisks whose file allocation maps are also small. Even in the base case, writing the entire map to DASD involved only minimal DASD I/Os. However, as the number of minidisk cylinders increases, the benefit of the performance improvement can be realized. The largest benefits will be seen when there is a high rate of file activity on very large minidisks.