This section shows the capability of PowerMax to maintain excellent performance for Oracle workloads, with or without storage group compression.
Recall from the PowerMax compression and deduplication section that the Adaptive Compression Engine (ACE) does not compress the most active data extents immediately, even if they belong to a storage group marked for compression. It keeps the most active 20 percent of the allocated storage capacity uncompressed (while storage space permits). Typically, the most recent data is being accessed most frequently. Over time, new data is written and accessed frequently. What was previously considered “hot” becomes less active and is automatically compressed.
While this method applies to real world database access patterns, benchmark tools tend to ignore it and run randomly across the whole database. The SLOB “hot-spot” feature allows access to a portion of each user table more frequently, simulating real world behavior.
To make the PowerMax compressions tests as realistic as possible, we loaded SLOB with semi-random data, resulting in a 3.0:1 compression ratio. We used a 5 GB buffer cache and SLOB hot-spot. This configuration resulted in a workload with 80 percent storage read I/Os and 60 percent cache read-hit. Therefore, 80 percent of the I/O requests sent to storage were reads, creating an OLTP type workload while also ensuring there were many requests for data that might be compressed. The 40 percent read-miss meant that of all the reads, at least 40 percent of the data was not found in the PowerMax cache and had to be brought from the flash media (compressed or not).
The SLOB workload was run using the two Dell servers in the cluster.
The following figure shows the compression-disabled test results taken from Oracle AWR. Under ‘Top Timed Events’, AWR reported a 0.28ms data file read latency (db file sequential read metric). Under ‘System Statistics (Global)’, the total data file IOPS were 253,477 (184,270 reads + 69,207 writes).
Figure 19. Storage group Compression disabled, AWR statistics
The following figure shows compression-enabled test results taken from Oracle AWR. Under ‘Top Timed Events’, AWR reports a 0.31ms data file read latency (the db file sequential read metric). Under ‘System Statistics (Global)’, the total data file IOPS were 250,743 (181,296 reads + 69,447 writes).
The two AWR reports for the PowerMax storage system with enabled and disabled compression show an approximate 1 percent difference in Oracle data file total IOPS and a 0.03 ms data file read response time difference. Users do not perceive these differences, which demonstrates the strength of the PowerMax architecture that supports data reduction while maintaining high performance.
Figure 20. Storage group Compression enabled, Top Timed Event