Performance Consistency with FlashArray NVMe-oF and SQL Server Linux

When it comes to database performance administrators don’t like surprises. Using the Pure Storage FlashArray™ with NVMe-oF transport protocol database administrators can achieve consistent performance levels for SQL Server workloads, and reduce the complexity of traditional DAS (direct attached storage) models.

NVMe-oF

The Pure Storage FlashArray with NVMe-oF delivers high performance in SQL Server environments providing steady write latency and bandwidth.  Our tests using a real world SQL Server OLTP database workload with realistic data consisting of average 10K writes, and a read/write ratio of 20/80 resulted in tight standard deviation levels.

Write latency standard deviation resulted in 6 microseconds after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration:

Write latency standard deviation resulted in 6 microseconds after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration
Write latency standard deviation resulted in 6 microseconds after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration

Write bandwidth standard deviation resulted in 6MB/s after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration:

Write bandwidth standard deviation resulted in 6MB/s after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration
Write bandwidth standard deviation resulted in 6MB/s after 30 minute ramp time measured at the FlashArray storage layer 8 hour duration

Test Drive FlashBlade

NVMe-oF Direct Attached Storage Headaches

On the surface NVMe-oF DAS (direct attached storage) sounds simple (i.e., just install some NVMe drives in a server) but looking deeper into the solution raises many questions.

DAS Drive Thermal Consideration

  • DAS drive/cards have different thermal capabilities, so which drive/vendor should you choose?

DAS Drive Performance Considerations:

  • Some DAS drive/cards will throttle performance for various reasons like heat levels, so how much and under what conditions?
  • DAS drive/cards caching capabilities are different,so how will that affect your SQL Server workload performance (cache flush, cache thrashing etc.)?

DAS Drive Support Consideration:

  • How is support handled between the DAS drive/cards, OS, and Server vendor?
  • Do you have to worry about blame shifting when troubleshooting issues?

DAS Drive Firmware Update Considerations:

  • How easy is it to update DAS drive/cards firmware for multiple drives across multiple servers? Is this process manual or automatic?
  • What is the firmware update notification process like?
  • For example, is it via the operating system, server, or drive vendor, and are these various update notification sources coordinating updates for you.

Server Thermal Considerations:

  •  Servers also have different thermal capabilities with regard to airflow and spacing that can be challenging to match the DAS drive/cards.
  •  Some servers have DAS drive/cards certification, but what if the DAS drive/cards you want to use are not certified with your server vendor?

NVMe DAS Resiliency Consideration:

  • NVMe-oF DAS drive/card RAID controller card support is few and far between, it is a somewhat of a  ‘wild west’ environment.

Summary

With the Pure Storage Flash Array you are getting an end-to-end engineered solution built with our DFM (Direct Flash Modules) and Purity operating environment.  Updates are simplified performance is consistent and efficient.

Trying to match DAS drive/card components with servers has so many performance environment unknowns.  Now you no longer have to rely on a generic DAS drive/cards vendor IO controller, instead with FlashArray DFM (Direct Flash Modules) we are the IO controller.