Release Notes

24 Dell PS Series Snapshots and Clones: Best Practices and Sizing Guidelines | BP1027
another must be allocated to the snapshot volume (or reserve). However, when a random I/O pattern is used,
additional pages are allocated for reserve more quickly as I/O’s are not always adjacent and may modify a
different page each time. As each subsequent I/O occurs to pages that have already been allocated, the rate
of consumption slows and ultimately levels off.
It should be noted that not all applications access data in purely random or purely sequential fashion and
many do not access 100% of the entire volume space. In fact, many applications access specific areas of
data more often than others. The results of this test only demonstrate the behavior of simulated access
patterns and do not represent actual customer results.
6.2 Reserve usage after cumulative snapshots
Many companies use snapshot technology to improve recovery time when compared to daily tape backups.
Snapshots allow users to take point-in-time copies of a volume. Users can create schedules to take multiple
snapshots per day of a single volume which provides the ability to recover a corrupted volume or a lost file
more quickly than from tape. However, having multiple point-in-time copies can impact the amount of storage
space utilized by the snapshot reserve. Moreover, as seen in section 6.1, having random or sequential I/O
access patterns makes a difference in how quickly snapshot reserve is consumed by an application.
Using the same pool configuration as in previous tests, multiple snapshots of a 10 GB volume were taken
over time while observing the snapshot reserve usage. However, in this test case the amount of the base
volume accessed was varied. This was done because different applications may access large or small data
sets and within those data sets, may access all of the data or just portions of it. For example, a user has an
application running and has a schedule to take hourly snapshots of the volume. In between these hourly
snapshots, the application is making changes to a percentage of the base volume. This test shows the impact
of this behavior on the snapshot reserve space. By varying the number of sectors in the Maximum Disk Size
value for IOmeter, only 10% of the base volume was initially accessed. Then, the access was increased in
steps of 10% until 50% of the volume was accessed. Each time the access size was increased; there was
another snapshot of the volume and the I/O workload was run. These tests were repeated for both random
and sequential access patterns.