PureStorage has 150TB DirectFlash Modules

I have just realized that PureStorage has 150TB DirectFlash Modules. 

That got me thinking. 

Flash capacity is increasing year by year. What are performance/capacity ratios?

The reason I'm thinking about it is that poor Tech Designer (like me) need some rule-of-thumb numbers for capacity/performance planning and sizing.

For example, back in the day, EMC used the rule of thumb that a 

• 300 GB 15k RPM SCSI/SAS drive can achieve 180 IOPS => 0.6 IOPS per 1 GB
• 300 GB 10k RPM SCSI/SAS drive can achieve 150 IOPS => 0.5 IOPS per 1 GB
• 1 TB GB 7.2k RPM NL-SAS drive can achieve 80 IOPS => 0.1 IOPS per 1 GB

With rotational disks, the performance was the issue, so we needed to start with performance sizing followed by capacity sizing. Then Flash was invented, and it switched the problem. There was a lot of performance (IOPS), but capacity was the challenge.

Now, when the capacity of flash is higher and higher, I'm always curious what is the IOPS per GB ratio of particular Flash Module or AllFlash storage system.

Let's show it in example of VMware vSAN and 6.4 TB module

VMware vSAN ESA is AllFlash (NVMe) software-driven storage system. Details about storage performance testing are available here. I was using 6.4TB 2.5in U.2 P5620 NVMe High Perf High Endurance flash disk. 

In P5620 spec sheet was written that it can achieve 

• Random Read (100% Span) - 1100000 IOPS (4K Blocks)
• Random Write (100% Span) - 390000 IOPS (4K Blocks)

For capacity planning and sizing, I usually use 32 KB IO size, which is the typical average in an enterprise data center, therefore, if we divide the above values by 8 (this is tricky, but let's simplify it), we would get 

• Random Read (100% Span) - 137500 IOPS (32K Blocks)
• Random Write (100% Span) - 48750 IOPS (32K Blocks)

And this is the backend performance. When there is storage magic (erasure codings, compression, dedup, etc.), the real usable performance is even lower. When I used the above 6.4 TB drive in vSAN ESA with RAID-5 with compression enabled, I was able to achieve the following usable performance

• 100% read I/O: 24,000 IOPS @ 32K, 2 ms latency
• 100% write I/O: 9,000 IOPS @ 32K, 2 ms latency

... and usable performance is what matters, right?

So now (after the real test), I know that 6.4TB 2.5in U.2 P5620 NVMe High Perf High Endurance flash disk in reality has 5.8 TB usable capacity in vSAN ESA, therefore, I see the following ratios

• 4 read IOPS per 1 GB
• 1.5 write IOPS per 1 GB

... and that's what I can use for capacity/performance planning and sizing, but every flash module is different and we do not have time to test everything. Do you have in PureStorage any sizer which would help technical designers with capacity/performance planning?

What is performance/capacity ratio of 150 TB module

And that is what I'm curious and I asked PureStorage Engineers (Jacint Juhazs and Pavel Kovar) on LinkedIn if they have some performance/capacity sizer for poor Technical Designers or they can share some performance results. I know, that storage vendors typically do not want to share performance results because performance is always tricky, but let's try to ask.

If Jacint or Pavel share some info, I will update this blog post.