System Dump Improvements
Abstract
Because z/VM 6.3 increased its supported storage to 1 TB, the z/VM dump program needed to add the capability to dump 1 TB. As part of these extensions, the rate at which dumps could be written to the dump devices was improved. For ECKD devices, dump rates improved 50% to 90% over z/VM 6.2. For EDEV devices, dump rates improved 190% to 1500% over z/VM 6.2.
Introduction
This article addresses performance enhancements for z/VM dump creation.
For ECKD, channel programs were improved to chain as many contiguous or noncontiguous frames as possible for each I/O operation.
For EDEV, each I/O now writes as many contiguous frames as possible.
These enhancements apply to both SNAPDUMP and PSW RESTART dumps.
Method
Virtual Storage Exerciser was used to create specialized workloads for evaluation of these performance improvements. The workloads merely populated a specific amount of storage prior to issuing the specific z/VM dump command being evaluated. Variations included number of guests and guest virtual storage size. Variations in these parameters affect the number of frames dumped by the z/VM dump program. Once the application had created the desired storage conditions, the appropriate z/VM dump command was issued.
Table 1 defines specific parameters for four load configurations. Results tables found later in this chapter identify loads by load configuration number defined here.
|
Table 1. Load Configurations |
||||
| Load Configuration -> | 1 | 2 | 3 | 4 |
| System Model | 2817-744 | 2817-744 | 2817-744 | 2817-744 |
| SYSGEN Storage | 256 GB | 256 GB | 1 TB | 1 TB |
| Processors (dedicated) | 4 | 4 | 4 | 4 |
| VIRSTOR Users | 1 | 10 | 10 | 1 |
| VIRSTOR End Addr | 230 GB | 3 GB | 100 GB | 990 GB |
| VIRSTOR Increment | 4 KB | 4 KB | 4 KB | 4 KB |
| Notes: 2817-744. I/O configuration details are included in each results table. | ||||
Experiments were constructed by varying configuration choices as follows:
- Both SNAPDUMP and PSW RESTART dumps were evaluated.
- Both ECKD and EDEV (Emulated FBA on SCSI) devices were evaluated.
- Storage sizes of 256 GB and 1 TB were evaluated.
- Measurements of 256 GB systems were completed on both z/VM 6.2 and z/VM 6.3.
- Measurements of 1 TB systems were completed only on z/VM 6.3.
The z/VM dump program collected the information necessary to determine the rate.
Results and Discussion
The results tables include the following columns:
- %FT, which is the percentage of dumped frames that are CP frame table frames.
- Elps Time, which is the dump elapsed time, in seconds.
- Rate, which is the dump rate, in frames per second.
256 GB Storage With ECKD Devices
Table 2 contains the results for both z/VM 6.2 and z/VM 6.3 measurements using ECKD devices in 256 GB of storage.
|
Table 2. ECKD Dumps With 256 GB |
|||||||||
| LOAD | DASD | Dump | Dumped | % | Elps | ||||
| Run ID | Conf | Type | Type | Rel | Size | Frames | FT | Time | Rate |
| JW1E2561 | 1 | ECKD | SNAP | 6.2 | 256 GB | 1013470 | 51 | 257 | 3943 |
| JX1E2563 | 1 | ECKD | SNAP | 6.3 | 256 GB | 1014243 | 51 | 157 | 6460 |
| JW1E2562 | 1 | ECKD | PSW | 6.2 | 256 GB | 1013612 | 51 | 248 | 4087 |
| JX1E2564 | 1 | ECKD | PSW | 6.3 | 256 GB | 1015453 | 51 | 159 | 6386 |
| JWAE2561 | 2 | ECKD | SNAP | 6.2 | 256 GB | 565281 | 92 | 178 | 3176 |
| JXAE2563 | 2 | ECKD | SNAP | 6.3 | 256 GB | 566038 | 92 | 92 | 6153 |
| JWAE2562 | 2 | ECKD | PSW | 6.2 | 256 GB | 565617 | 92 | 178 | 3178 |
| JXAE2564 | 2 | ECKD | PSW | 6.3 | 256 GB | 566058 | 92 | 92 | 6153 |
| Notes: 2817-744. Four 8.0 Gbps Fibre-channel switched channels, 2107-E8 control unit, 14 3390-3 volumes. | |||||||||
Results from this environment demonstrate a 50% to 90% dump rate improvement over z/VM 6.2 for both SNAPDUMP and PSW RESTART dumps.
256 GB Storage With EDEV Devices
Table 3 contains the results for both z/VM 6.2 and z/VM 6.3 measurements using EDEV devices in 256 GB of storage.
|
Table 3. EDEV Dumps With 256 GB |
|||||||||
| LOAD | DASD | Dump | Dumped | % | Elps | ||||
| Run ID | Conf | Type | Type | Rel | Size | Frames | FT | Time | Rate |
| JW1S2561 | 1 | EDEV | SNAP | 6.2 | 256 GB | 1025710 | 51 | 903 | 1136 |
| JX1S2565 | 1 | EDEV | SNAP | 6.3 | 256 GB | 1026690 | 51 | 308 | 3333 |
| JW1S2562 | 1 | EDEV | PSW | 6.2 | 256 GB | 1030687 | 50 | 909 | 1134 |
| JX1S2566 | 1 | EDEV | PSW | 6.3 | 256 GB | 1026735 | 51 | 308 | 3334 |
| JWAS2561 | 2 | EDEV | SNAP | 6.2 | 256 GB | 577397 | 90 | 638 | 905 |
| JXAS2566 | 2 | EDEV | SNAP | 6.3 | 256 GB | 578493 | 90 | 38 | 15224 |
| JWAS2562 | 2 | EDEV | PSW | 6.2 | 256 GB | 577599 | 90 | 638 | 905 |
| JXAS2567 | 2 | EDEV | PSW | 6.3 | 256 GB | 578258 | 90 | 38 | 15217 |
| Notes: 2817-744. One 16.8 Gbps Fibre-channel-protocol channel, 6310-80 control unit, 100 9336 volumes. | |||||||||
Results from this environment demonstrate a 190% to 1500% dump rate improvement over z/VM 6.2 for both SNAPDUMP and PSW RESTART dumps. The dump rate to EDEV devices is highly dependent on the number of contiguous frames. The best improvements occurred for load configurations where the dumped frames were 90% CP frame table frames.
1 TB Storage With ECKD Devices
Table 4 contains the results for z/VM 6.3 measurements using ECKD devices in 1 TB of storage.
|
Table 4. ECKD Dumps With 1 TB |
|||||||||
| LOAD | DASD | Dump | Dumped | % | Elps | ||||
| Run ID | Conf | Type | Type | Rel | Size | Frames | FT | Time | Rate |
| JXAE1T4 | 3 | ECKD | SNAP | 6.3 | 1 TB | 4138337 | 50 | 639 | 6476 |
| JXAE1T5 | 3 | ECKD | PSW | 6.3 | 1 TB | 4138299 | 50 | 631 | 6558 |
| JX1E1T3 | 4 | ECKD | SNAP | 6.3 | 1 TB | 4154508 | 50 | 632 | 6574 |
| JX1E1T4 | 4 | ECKD | PSW | 6.3 | 1 TB | 4154470 | 50 | 632 | 6574 |
| Notes: 2817-744. Four 8.0 Gbps Fibre-channel switched channels, 2107-E8 control unit, 14 3390-3 volumes. | |||||||||
Results from this environment demonstrate dump rates within 10% of all dump rates seen in the 256 GB measurements.
1 TB Storage With EDEV Devices
Table 5 contains the results for z/VM 6.3 measurements using EDEV devices in 256 GB of storage.
|
Table 5. EDEV Dumps With 1 TB |
|||||||||
| LOAD | DASD | Dump | Dumped | % | Elps | ||||
| Run ID | Conf | Type | Type | Rel | Size | Frames | FT | Time | Rate |
| JXAS1T6 | 3 | EDEV | SNAP | 6.3 | 1 TB | 4150681 | 50 | 1308 | 3173 |
| JXAS1T7 | 3 | EDEV | PSW | 6.3 | 1 TB | 4150648 | 50 | 1305 | 3181 |
| Notes: 2817-744. One 16.8 Gbps Fibre-channel-protocol channel, 6310-80 control unit, 100 9336 volumes. | |||||||||
Results from this environment demonstrate dump rates equivalent to those seen in the 256 GB measurements.
Summary and Conclusions
For ECKD devices, dump rates improved 50% to 90% over z/VM 6.2.
For EDEV devices, dump rates improved 190% to 1500% over z/VM 6.2.
For ECKD devices at 1 TB, all dump rates were within 10% of all dump rates achieved at 256 GB.
For EDEV devices at 1 TB, dump rates were within 10% of all dump rates achieved at 256 GB for similar load configurations.