We used HammerDB 3.3 to run the OLTP test case. This test case demonstrates a realistic workload with a high degree of locality of reference where customers access the most recent data frequently. The data is partially cached in PowerMax and 70% of the data reads are serviced from the PowerMax cache. We focused on achieving high IOPS while maintaining relatively low latencies.
The table below shows the Microsoft SQL Server iSCSI tests using a HammerDB OLTP benchmark. The database was configured with 10,000 warehouses. We used different numbers of warehouses during the tests to demonstrate the iSCSI performance and impact of high degree of locality of reference and PowerMax cache for those workloads. As mentioned earlier we were limited by the number of network paths available for our tests, but we still achieved very low latency and high SQL Server transaction rates for all these tests. At a cache hit rate of 49% and 10,000 warehouses, our test produced host IOPS of 227K with 77% reads and 23% writes. As we increased the cache hit rates by changing the number of warehouses, our test produced a read response time of 0.35 ms and a write response time of 0.42 ms while achieving 216K host IOPS for a 2,500-warehouse test.
Figure 20. Microsoft SQL Server iSCSI test results
It is evident that with a realistic workload, our lab environment featuring a single-brick PowerMax was able to consistently generate close to a quarter million IOPS with sub-millisecond read and write latencies. Although we were limited by our network bandwidth, and by the number of ports and paths available for iSCSI traffic, we achieved low latencies and higher performance while maintaining low system utilization. We can even achieve higher IOPS and maintain low latencies by scaling up the system by adding PowerMax bricks and iSCSI front-end ports.