IBM Boosts Capacity On FlashSystem Arrays

March 2, 2015 Timothy Prickett Morgan

If you are looking to shop for all-flash storage systems to attach to your IBM i or other systems in the datacenter, you might want to take a gander at the new FlashSystem arrays that Big Blue launched last week. IBM has partnered with memory manufacturer Micron Technology and moved from the enterprise multi-layer cell (eMLC) memory used in the FlashSystem 840 and V840 machines to more capacious and presumably less expensive NAND MLC memory in the new Flash System 900 and V9000 machines launched last week.

From the looks of things, the basic feeds and speeds of the FlashSystem controllers look about the same as those used in the FlashSystem 840 and V840, and it is a bit perplexing why the new machines are called the 900 and V9000. The V in the product name means that the access to the flash is virtualized by pairing it with IBM’s Storage Virtualization Controller (SVC), and specifically, IBM puts a pair of these controllers onto a single flash array to make it virtualized. The SVC has snapshotting, compression, and other features that were not part of the original FlashSystem controllers. Why the V9000 has an extra zero in its name compared to the 900 is anyone’s guess; it could be an expensive typo, but there was probably some logic to it somewhere in the IBM Marketeering Department.

The big change is with the underlying flash, which Micro calls FortisFlash and which is a new MLC NAND flash that is less expensive per unit of capacity than the eMLC flash that IBM had been using. Working in collaboration with Micron, IBM tweaked the FlashCore controller at the heart of the FlashSystem to provide long life, high duty cycle, and low latency access to data. The upshot is that the FlashSystem 900 and V9000 arrays provide similar or better performance and higher capacity. (More on pricing in a moment.)

The FlashSystem 900 has flash modules that come in 1.2 TB, 2.9 TB, and 5.7 TB capacities, and up to a dozen of them plug into the front of the 2U chassis. With RAID 5 data protection turned on across the flash modules, you get from 2.4 TB of capacity for four modules up to 57 TB of capacity for 12 modules. That is a 40 percent boost in capacity in the same form factor. The minimum write latency is 155 microseconds and the minimum read latency is 90 microseconds, showing once again the disparity between reads and writes that is inherent in flash. A fully loaded FlashSystem 900 can deliver 1.1 million I/O operations per second on 100 percent random reads and about 600,000 IOPS on 100 percent random reads; with a mix of 70 percent reads and 30 percent writes, the FlashSystem 900 can do about 800,000 IOPS. (That is with 4 KB file sizes, a common size for storage benchmarks.) In terms of bandwidth, with 256 KB file sizes, the machine can pump out 10 GB/sec doing reads and 4.5 GB/sec doing writes. The FlashSystem 900 has AES-XTS 256 encryption built into its redundant controllers, which are hot swappable. The array can link to servers with Fibre Channel links running at 8 Gb/sec and 16 Gb/sec, iSCSI links running at 10 Gb/sec, and InfiniBand links running at 40 Gb/sec and 56 Gb/sec. (The spec sheets don’t mention the higher speed InfiniBand, but the price list does.)

The FlashSystem 900 on top, and the FlashSystem V9000 on the bottom

With the FlashSystem V9000, IBM puts an SVC controller on the top and bottom of the FlashSystem 900 and puts it all inside a 6U chassis. The SVCs each have a single eight-core Xeon E5-2600 v2 processor and 64 GB of memory to run the SVC software. The FlashSystem V9000 comes with 1.3 TB, 2.9 TB, and 5.7 TB MLC drives from Micron. (Why it is 1.3 TB instead of 1.2 TB with the FlashSystem 900 is not clear.) Within a single device, The FlashSystem V9000 delivers up to 57 TB of capacity using the fattest of the Micron MLC NAND modules, and by hanging extra disk controllers off it, the capacity can be scaled to up to 2.28 PB of effective capacity in a single system with compression on. (IBM says it can get 5 to 1 compression on typical data.) With the SVC’s data compression turned off and in scale-out mode, the FlashSystem V9000 has a latency of 200 microseconds on reads with 4 KB files and can push 2.52 million IOPS. If you turn on the in-line data compression from the SVC, it drops to 1.2 million IOPS, but you can cram up to five times as much data onto the array. IBM did not provide write performance for the FlashSystem V9000 with either compression on or off.

Brace yourself for the prices if you need this kind of performance out of your storage. The base FlashSystem 900 costs $32,000 and a 1.2 TB flash module costs $17,350, with the 2.9 TB module costing $37,400 and the 5.7 TB module costing $74,800. The encryption module costs $15,000, and the host adapters cost anywhere from $3,500 to $6,500 a pop. That works out to somewhere around $960,000 fully loaded at list price for the FlashSystem 900. To turn this into a FlashSystem V9000, it costs another $27,500 to wrap the control enclosure around the FlashSystem 900. You build it out by adding more enclosures from there, and the pricing is not obvious to me unless you just add multiple FlashSystem 900 arrays to the initial V9000 machine to expand it. The FlashSystem V9000 can scale to four flash modules, and another four can be added to this for a total of eight. That is 456 TB of raw capacity, and with compression it expands to 2.28 PB in 34U of rack space. It is not at all clear what the performance on reads and writes would be on such a fat FlashSystem. But what is clear is that it would cost many millions of dollars.

One thought on “IBM Boosts Capacity On FlashSystem Arrays”

Spectrum Scale vs. WekaIO says:

September 24, 2018 at 2:44 pm

[…] storage “devices” used in the test are high end Texas Memory Systems arrays that cost over $1 Million. Given that, I think Big Blue could have afforded a few more servers to push the load. More […]

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search