Release Notes

23 Dell PS Series Snapshots and Clones: Best Practices and Sizing Guidelines | BP1027
6 Test results and analysis
This section details the various test scenarios performed and explains the results of each test.
6.1 Effect of block size and random vs. sequential I/O pattern
Data warehouse and business intelligence applications typically have a higher percentage of sequential I/O
operations as well as a higher percentage of disk reads. Online transaction processing (OLTP) workloads are
typically more random in nature and disk writes may be 30-50% of the total I/O operations. Although each
application may have a unique I/O pattern, the effects of purely random and sequential I/O access patterns on
the snapshot reserve space were tested to demonstrate their effects on snapshot reserve.
In order to simulate such an environment, three PS6010XV arrays were configured in a single RAID 10 pool.
A 100 GB volume was created and attached to a Windows 2008 R2 host. After a snapshot of the test volume
was created, IOmeter was used to run a random I/O workload to the base volume (as the snapshot volume
remained offline) while the snapshot reserve usage was monitored at regular intervals. After collecting the
results, the snapshot was deleted and the test repeated with a sequential workload.
For both workloads, the read/write mix was 67% reads and 33% writes with one worker and two outstanding
I/O’s accessing 100% of the volume. The intent was only to provide a steady state workload – not to test any
maximum capability of the system.
Reserve utilization with evenly distributed I/O
The results show that I/O access patterns will make a difference in how quickly snapshot reserve is
consumed by an application. A PS Series array will allocate virtual pages of storage to a volume when it is
created or expanded. The same is true for the snapshot reserve and the diagram illustrates, a purely
sequential access to a volume takes longer, because it requires more I/O operations, to fill up a page before