This is the last post in this series and I’ll not introduce anything new here, but rather just summarise the changes explained and talk a bit about the value the solution delivers to the organisation.
Let’s first review the situation we faced before implementing the changes.
The cost of writing the log-records to the database was that all the parallel writing from many different sources was such that it introduced severe bottlenecks to the point that the logging feature had to be turned off days at a time. This was not acceptable but rather than shutting down the whole system which would put lives in immediate danger, this was the only option available. Then even if that would have been fast enough, the moving of data was taking over twice the time available and it was fast approaching the point where data written in 24 hours would take more time to move to the historical store for log-data. That would of course have resulted in an ever growing backlog even if the data move was on 24×7. On top of that the data took up 1.5 TB of disk space, costing a lot of money and raising concerns with out ability to move it to EXADATA.
To resolve the issue during business hours of having contention causing a crippling impact on the overall system, we changed the table setup to not have any primary keys, no foreign keys and no indexes. We made the tables partitioned such that we get one partition per day.
To make the move from operational tables to historical tables faster, we opted to have both in the same instance on EXADATA. This allowed us to use partition exchange to swap out the partition from the operational table and swap it into the historical table. This took just a second as all we did was updating some metadata for which table the partition belongs to. Note that this ultra fast operation replaced a process that used to take around 16 hours, for which we had 6.5 and the time it took was expanding as business was growing.
Finally, to reduce the space consumed on disk we used HCC – Hybrid Columnar Compression. This is an EXADATA only feature for compressing data such that columns with repeating values gets a very good compression ratio. We went from 1.5 TB to just over 100 GB. This means that even with no purging of data it would take us over five years to even get back to the amount of storage this used to require.
So in summary
- During business hours we use 20% of the computing power and even less of the wall clock time it used to take,
- The time to move data to the historical store was reduced from around 16 hours to less than one second.
- Disk space requirement was reduced from 1.5 TB to just over 100 GB.
And all of this was done without changing one line of code, in fact there was no rebuild, no configuration change or anything to allow this drastic improvement to work with all the different systems that was writing to these log-tables.
One more thing to point out here is that all these changes was done without using traditional SQL. The fact that it is an RDBMS does not mean that we have to use SQL to resolve every problem. In fact, SQL is often not the best tool for the job. It is also worth to note that these kinds of optimisations cannot be done by an ORM, it is not what they do. This is what your performance or database architect needs to do for you.
For easy lookup, here are links to the posts in this series.