Pure’s all-flash storage platforms: engineered for sustainability  

SPONSORED FEATURE: Highly sustainable IT systems are better for the long term health of our planet because they contribute less to greenhouse gas emissions and global warming. They have systems and devices with lower amounts of embedded carbon, which use less electricity when in use and result in less e-waste because they are used for longer and have greater recyclability.

Pure Storage has adopted sustainable storage system design because it is good for the environment and because customers are increasing their focus on Environmental, Social and Governance (ESG) issues when selecting IT systems.

Alex McMullan.

We interviewed Alex McMullan, CTO International at Pure Storage to find out more about its sustainability tenets, splitting our talk into a discussion at the device level and then broadening it out to systems and operational use.

The company does not build its FlashArray and FlashBlade systems using conventional SSDs with their on-board controllers looking after flash management, IO operations, wear levelling and drive health. Instead it builds its own Direct Flash Modules (DFMs) from NAND chips and its Purity operating system looks after flash management, IO operations, wear levelling and NAND health from a system point of view instead of a drive point of view.

 Electricity usage

We asked him if there was any way he could compare the 36 TB DFM’s electricity usage to equivalent capacity SSDs.

He said there were three primary dimensions to such a comparison: ”The first of those is, obviously, the NAND capacity itself. The way in which we manage and maintain wear at a system level, not at the drive level, means we don’t have to over-provision as much inside the device. Our over-provisioning is a very, very small percentage; it’s a single digit number where traditionally, our competitors will have over-provisioned in the 20 percent range.”

A 36TB conventional SSD with 20 percent over-provisioning will actually contain 43.2 TB of raw NAND, while a 36 TB Pure DFM with less than 5 percent over-provisioning will have, say, 37.5 TB of raw flash. A 1 PB conventional all-flash array built from 28 such drives will actually have 1.21 PB of raw NAND, 210 TB extra, which will draw electricity. 

The equivalent Pure system will have, say, 1.05 PB of raw flash, which will need less electricity.

McMullan said: “It’s simply a reflection on efficiency and being able to manage wear at the system level rather than on an individual device. That’s the first dimension in terms of having to put less components in which then obviously draws less power as a result.”

 He quantified this, saying: “The Solidigm 61.4 TB SSD actually does a very good job in terms of power management, but still, its maximum power is 25 watts per device. Our 75 TB DFM is going to draw less than 10 watts.”

On that basis our hypothetical 1PB storage chassis filled with 17 x 61.44 TB Solidigm drives will draw up to 407.5 watts while an equivalent 1PB chassis filled with 14 x 75TB DFMs will draw up to around 126 watts. Over 12 months that 281.5 watt difference will have a financial impact. It will also have an effect on a datacenter’s power budget with less used up by NAND capacity  composed of Pure DFMs than in this case by Solidigm SSDs. Keep this point in mind.

Also, McMullan said Pure does not intend to increase the DFM power budget when moving from 75 TB to 150 TB DFMs.

NAND vs Disk

How about comparing a Pure DFM array to a disk drive array? McMullan said:” The HDD-based systems have obviously the dual challenge of the moving actuator(s) and moving platters. So one of those is fixed in terms of power consumption. The other is variable in terms of reading and writing. So we observe HDD-based systems are much more subject to workload in terms of power draw than solid state based systems.”

Without cherry picking statistics, and using the latest HDD vendor datasheets: “How we’re positioning for now is eight TB/watt compared to between one and two TB/watt” for the disk drive array.

The disk drive array would take up more rackspace than an equivalent capacity Pure DFM chassis, and put out more heat. This has an effect, with McMullan telling us: ”There’s a consequential input of more power and then it requires more chilling to extract that thermal capacity as well. … It’s multiplicative in the way in which more drive power-in requires more power to manage that input power so you’re getting the doubling effect. You need to have more chiller power to consume that heat.”

The DRAM aspect

McMullan’s second dimension? “DRAM is required inside any SSD, for the correct functioning of the NAND controller that’s embedded in there. .. The difference with the way we use them is that we don’t need our DFM to track LBA (Logical Block Addresses) as an indirection. It doesn’t need to do garbage collection, it doesn’t need to do that traditional SSD workload, so we don’t have to run it at a higher clock rate.”

“More interestingly, we don’t need anything like as much DRAM as a comparative size device. The Solidigm 61.44 TB SSD has, from their spec sheets, 20 GB of DRAM on board. And that’s probably down to, I would guess, 16 gigabytes after ECC protection on that memory.”

“Our currently shipping 75 TB DFM has considerably less DRAM. It’s the minimal ECC protective footprint that we can put in that device.”

For Pure there isn’t a linear relationship between DRAM need and drive NAND capacity: ”We anticipate the same amount of DRAM going into our 150 TB drive when that ships at some point in the future, as in the current 75 TB one. Making my point that we don’t need memory per terabyte per capacity.”

Why is that?

“Because our NAND controller is doing NAND management in terms of driving channels to the chip packages. It’s not doing all the other traditional SSD-based tasks that conventional SSD controllers have to do.”  Pure’s DFMs don’t do metadata management or garbage collection.  Both of those cause DRAM capacity requirements to grow linearly as SSD capacity increases.

And the third dimension? Super-capacitors.

 “Broadly, the supercaps are there to provide power protection for the contents of DRAM. Logically, the amount of capacitance you have to put on board one of our DFMs is much less” than with off-the-shelf SSDs because the DFMs have less DRAM than an equivalent capacity SSD.

All-in-all, McMullan reckons: “We’re leading the world in making bigger solid state devices, which can store more data in less footprint with less power.”

[Part two of this article; Sustainable systems for the Age of AI, follows here.]

Note. This article was sponsored by Pure Storage.