The automated tuning of the Automated QoS function is realized by adjusting bandwidth so that the volume in which the target is set may achieve a target. At this time, the bandwidth of the volume in which the target is not set is adjusted automatically at the same time.
Evaluation is performed based on actual measured Read + Write response time for the volume.
Furthermore, tuning is performed in the range of hardware marginal performance. The set target value is not ensured.
Automatic Tuning Example
Implementation example of automatic tuning is described below.
When it is determined from the performance of tasks that the performance of volume 1 (Vol#1) is insufficient, as the actual measured value for the response time is 50 msec, the user, in order to make this faster than 50 msec, configures a target value of 30 msec. Target values will not be configured to the other volumes (Vol#2, Vol#3).
As a result, the bandwidth of Vol#1 expands, the bandwidths of volume 2 (Vol#2) and volume 3 (Vol#3) are narrowed, and automatic tuning aimed at the target value is performed.
As a result of the adjustments of bandwidths, the response time of Vol#1 reaches the target.
Figure 5.2 Structure of Bandwidth Adjustments
Structure of Automatic Tuning
The QoS function of ETERNUS Disk storage system sets the bandwidth limiting by 16 levels from 60IOPS and 5MB/s to unrestricted. The Automated QoS function adjusts the performance by changing bandwidth by using this function.
Immediately after a target value is set to a target volume for the first time, the performance is tuned from the initial value step by step, thus the effectiveness may not be seen soon.
Automatic tuning is realized by retrieving performance information at every evaluation interval (1 minute) and repeating the following processes.
The shared resource (CM, Port, switch Port, pool) having the highest Busy-rate is located.
The bandwidth is adjusted between the volumes sharing the resource having the highest Busy-rate. The details of the adjustment are as follows:
The bandwidths of volumes of which actual measured performance exceed the targeted performance are narrowed by one level.
The bandwidths of volumes with actual measured performance lower than the targeted performance and subject to narrowed bandwidths due to QoS are expanded by one level.
Adjustments for the shortages occurring from the above are made to the volumes to which target performance values have not been configured.
Note
The Automated QoS function references the performance information of the storages to which the FTV subject to performance adjustments belong, and the performance information of the switches.
Although the Automated QoS function can be utilized even when the performance management function of the switches are disabled, in such case, shared resources related to CM and pools will be the only subjects of performance adjustments.
Consideration of System in Low Load State
In business operation, the state of load differing depending on the time of the day can be considered.
When being in a low load state, the actual measured response time will be the fastest since throughput is low. As a result, the actual measured response time will exceed the target response time, and the bandwidth thereof will be in its narrowest state due to automatic tuning. From this state, when the time of the day of a high load state is entered and throughput increases, there is the possibility of response time worsening in an extreme manner.
When the actual measured response time of a volume to which a target value has been configured exceeds the target response time, only when the actual measured response time of another volume, to which a target value have been configured and is subordinate to a shared resource with a high Busy-rate, is lower than the target response time, the bandwidth is narrowed. When all volumes subordinate to the shared resource exceed the target response time, the bandwidths are not adjusted.
Even if there is no load on a volume from business operation, the bandwidth of the volume is not adjusted.
In addition, if there is no workload and there is some spare space in bandwidth, this bandwidth is temporarily supplied to other volumes short of bandwidth to cover shortfall. In this way, even when a target response time is satisfied, if there is spare space in bandwidth on any other volume, performance more than the target response time may be achieved.
