On February 11th/12th 2020 (depending which timezone you are in!) IBM announced that it was updating the entire modular entry, midrange and enterprise block portfolio. The family name of Storwize being replaced with the FlashSystem family.

For the 5000 range, this really is a name change, since we only updated the hardware during 2019, so the Storwize V5000 and V5100 become FlashSystem 5000 and 5100.

For the 7000 and 9000 series we introduced the next generation of NVMe controllers.

We also announced the second generation of Flash Core Modules (FCM) and the first set of Storage Class Memory drives that can be used in any NVMe capable Spectrum Virtualize based system.

Storwize to FlashSystem Decoder Ring

Old Product	New Product	Drives	Notes
Storwize V5010	FlashSystem 5010	SAS	Name change only
Storwize V5030	FlashSystem 5030	SAS	Name change only
Storwize V5100	FlashSystem 5100	NVMe	Name change only
Storwize V7000 Gen3	FlashSystem 7200	NVMe	New hardware (gen4?!)
FlashSystem 9110	n/a	NVMe	Discontinued
FlashSystem 9150	FlashSystem 9200	NVMe	New hardware (gen2?!)

FlashSystem 7200

The FlashSystem 7200 is an evolution of the previous generation Storwize V7000, with the same basic enclosure hardware (metalwork only) supporting up to 24 NVMe Flash drives or FCM, with up to four SCM drives per enclosure.

One of the main differences between the previous V7000 hardware is that all six PCIe slots can be populated with Host Interface Cards (HIC) – where previously two were always reserved for a SAS expansion attachment card. You therefore now have the option of still using a SAS expansion card if you wish to expand the capacity beyond what is supported by the base 24 slots.

However, with the second generation of FCM, having a maximum usable capacity (before data reduction) of 38.4 TB, the base enclosure can hold some 750TB of usable capcity, after RAID, and before data reduction. Add in the maximum FCM compression capability of 88TB per FCM, and that is over 1.5PB of effective usable capacity from hardware FCM compression alone.

Other notable updates are a base cache memory increase to 256GiB, upgradable in two stages to 1.5TiB. The CPU’s are updated to use Intel’s Cascade Lake family with 32 cores per enclosure. All PCIe slots are 16 lane Gen3 to allow full bandwidth capability to each Fibre Channel 32Gbit 4 port card. Up to 24x 32Gbit ports per enclosure.

FlashSystem 9200

As you probably can guess, the 9200 is an evolution of the 9150, and follows the same basic principals and specs of the 7200 with significantly more horsepower at hand.

The 9200 can also support up to 1.5TiB of cache, but provides 64 cores per enclosure allowing for even more system throughput, particularly when making use of the advanced functions provided by the embedded Spectrum Virtualize software.

Gen2 Flash Core Modules (FCM)

IBM is one of the few major storage controller vendors to build their own Flash drives. But FCM’s are much more than just your average Flash drive. Check out my post from 2018 when we first introduced FCMs for some of the basic differentiators.

The second generation builds on these computational storage devices by adding more AI infused smarts. Since the drives already monitor the heat of your data, using this with the health binning algorithms to extend the effective wear and lifetime of the base NAND flash, so these second generation use the heat to enable a dynamic SLC based cache algorithm to place your most frequently read data on the faster SLC like areas of the device. By using SLC caching, the endurance and performance of the device can be accelerated, think of this as pseudo EasyTier on the device itself. Tiering the data to provide the overall best latency to the upstream systems.

IBM has great plans for the future of FCM devices, with research and development working on many more industry unique and product defining differentiators over the next few years.

Storage Class Memory (SCM)

With the latest software updates IBM has now turned its “SCM ready” statements into a reality. Any of the NVMe based Spectrum Virtualize platforms can now support up to four SCM drives per enclosure. (only four may seem limited, but wait till you see the price and performance!)

In theory, the SCM drives can supply up to half a million IOPs per device – at latencies in to 10’s of microseconds (or less). We are still in the ‘million IOPs’ era, where every device out there is claiming millions of IOPs support, yet I see very few clients getting even close to that. Yes, there are some exceptions, but for the most part this is overkill.

The capacities at the moment are much smaller than the Flash drive equivalents and vary by the two different technologies. Intel’s Optane drives (based on their 3D-Xpoint technology built in colaboration with Micron). 3D-Xpoint really is a brand new technology, a sort of phase change style silicon technology that removes the need for transistors at each cell (as per DRAM) and so can reduce the cost and size needed but get much closer to DRAM latency than NAND ever will. The other technology is from Samsung, Z-NAND, in their Z-SSD drives which from what little I can find would seem to be a kind of revisiting of SLC Flash from 10+ years ago but with even better latency. Its not really a betamax/VHS race here, but to me it feels like 3D-Xpoint has more potential in the future and doesn’t suffer the same write penalty as NAND.

Latency is king however. Everyone wants lower latency. So lets assume that you can afford a couple of SCM drives, put them in a RAID-1 mirror and add to your storage pool made up of Flash drives, or better still FCMs. Now we can EasyTier between them and we see the same benefit we did 10 years ago with a small amount of Flash assisting the majority of HDD capacity.

Tiering is here to stay, and any vendor who tells you otherwise is covering up for the fact they don’t support tiering. The days of just an All-Flash array will soon be over, as we transition to SCM (and beyond) just as we have done from HDD to Flash in the last 10 years… exciting times.

---

We announced so much stuff last month that I’m still not done, next time we will talk about the new SVC platforms…