Faster I/O Subsystem Causing More WRITELOG Waits?

Paul Randal explains how WRITELOG waits can increase with a faster I/O subsystem:

I was contacted last week by someone who was asking how are faster disks causing more WRITELOG waits. They were seeing lots of these waits, with an average wait time of 18ms. The log was stored on a Raid-1 array, using locally-attached spinning disks in the server. They figured that by moving the log to Raid-1 array of SSDs, they’d reduce the WRITELOG wait time and get better workload throughput.

They did so and got better performance, but were very surprised to now see WRITELOG as the most frequent wait type on the server, even though the average wait time was less than 1ms, and asked me to explain…

Azure SQL DB Storage Bottleneck

If you are using Azure SQL Databases, you should definitely read this post by Brent Ozar: There’s a bottleneck in Azure SQL DB storage throughput. The bottom line:
the transaction log throughput currently appears to bottleneck at 16 cores!

The bit where he compares AWS costs/relative performance is also an eye opener:

  • 8 cores, 1,991 per month: 64 minutes</li><li><strong>16 cores,3,555 per month: 32 minutes (and interestingly, it’s the same speed with zone redundancy enabled)
  • 80 cores, 18,299 per month: 32 minutes</strong></li><li>Just for reference: 8-core AWS EC2 i3.2xl VM,1,424 per month with SQL Server Standard Edition licensing: 2 minutes (and I don’t put that in to tout AWS, I just happen to have most of my lab VMs there, so it was a quick comparison)