Client IO Performance Review Description of the physical hardware configuration including CPU, Memory, Disk, HBA,Storage Type, etc? Interview technical contacts to get a description of the problems. Measure IO using OS and Oracle tools available? Use R…

I look forward to your comments there. I have tried in that post to specifically say what optimizations are Oracle-specific.

It depends on the schema and the workload. For example, you might have a table such as the ‘Ontime Flightstats’ data, which is really a star schema but it is flattened into a single table.

This type of schema might benefit from throughput more than IOPS, particularly if covering indexes can be scanned sequentially. In fact, both MyISAM and InnoDB can take advantage of covering indexes to approximate the performance of a column store when scanning only the columns in the index. If MySQL has to get the whole row from the table, there is likely going to be a random element (since keys are not in natural table order in most cases) and then throughput can become less important than IOPS.

For tables with joins, if you use FORCE INDEX to force a scan of the fact table, and the dimension tables fit into memory, then throughput may be more important than IOPS. This is helped by a split LRU, since hopefully the scan of the fact table doesn’t evict the dimensions.

Oracle also supports parallel query for partitioned tables. This makes a tremendous difference in overall performance. If you have lots of throughput, Oracle can sequentially scan more than one partition in parallel, leading to serious performance improvements. My Google code project ‘Shard-Query’ shows that MySQL could be significantly faster if it supported this rather simple optimization.

Can MySQL use the sequential scanning more if it’s looking at MyISAM data? After all, MyISAM is sequential by nature; and the underlying [default] BTREE structure of data and indexes lends well to sequential scans….

Also, how does something like page size come into play? It’s not really changeable with MyISAM (unless you change the OS page size) but with InnoDB you can change the page size….

Because of nested loops, a MySQL DW will often have to perform many more random operations than a similar Oracle DB on the same schema/data. This is because Oracle can switch to a hash join, and sequentially scan large chunks of data. MySQL can not do this.