Standard values for the device workload

Disks (more precisely: logical volumes)

If volumes are shared by two or more users, the size of the workload is a key criterion for the wait time in front of each volume. With online applications, the wait time should not be more than one third of the hardware service time.

Therefore it is essential to determine whether the volume concerned is being shared by two or more users (or tasks) or whether it is assigned permanently to one user.

Magnetic tape cartridge units

Usually magnetic tape cartridge units are permanently assigned to a single task. Therefore, a workload of 100% is feasible.

The user should make sure here that the workload is not made artificially heavy by the “processing” of an excessive number of inter-block gaps.

ANALYZER reports for assessing device workload (volumes)

Report group	Report	Meaning
CATEGORY-IO	Duration of non paging IOs for category	Hardware and software service time for DMS I/O operations/s per task category
	Duration of paging IOs for category	Hardware and software service time for paging I/O operations/s per task category
DISK	IOs	Number of I/O operations per second and volume
	Queue length	Average length of queues for the individual volumes
	Data per IO	Number of PAM blocks per I/O operation and volume
	Time	Hardware and software service time for DMS I/O operations/s per volume
SERVICETIME	Duration of IOs for device	Hardware service time for I/O operations per volume with detailed breakdown and the wait time before the volume

Storage systems of type ETERNUS DX/AF can be monitored with the openSM2 Manager.

Tuning measures for device overloads (volumes)

The first thing to do is always to check the hardware service time (report “Time” of the report group DISK with the SAMPLING-DEVICE monitoring program switched on), taking into account the number of transferred PAM blocks per I/O operation.

In very rare cases the analysis must be continued as follows:

1. If the hardware service time is only negligibly longer than the DEVICE CONNECT TIME (report “Duration of IOs for device” of the SERVICETIME monitoring program or report group), you are dealing with read hits or fast write hits.

   The overloading of the logical volume is caused by the frequency of the accesses and leads to wait times for the tasks in front of the volume. In this case, it makes sense to use the PAV function. If you do not, it is necessary to relocate particularly frequently used files to other, less busy volumes (if possible also on a different physical disk drive).

2. If the hardware service time contains a significant share of DEVICE DISCONNECT TIME (report “Duration of IOs for device” of the report group SERVICETIME), the cause may be one of the following:

   • Insufficient support for the read caching due to too small a cache.

   • If there is a lack of “breathers” during continuous write loads, cache contents cannot be saved to disk in the interim.

   • A competition situation between logical volumes on the same disk drive (check the volumes per disk drive with the follwoing command of the SHC-OSD software product:

     /SHOW-STORAGE-DEVICE-CONFIG UNIT=*BY-VOLUME (...), INF=*PHYSICAL .

3. A share of REMAINING SERVICE TIME (report “Duration of IOs for device” of the report group SERVICETIME) in the hardware service time indicates the following:

   • Dilation of the hardware service time by other guest systems under VM2000. In this case, you should check the setting for the CPU quotas.
   • Use of REC / SRDF (see "Performance behavior in the case of synchronous remote replication" (Replication: volume-based mirroring for storage systems)).
     A REMAINING SERVICE TIME part, that is the same size as the DEVICE CONNECT TIME, indicates a number of “Remote Data Links” that is too low.