There’s been an ongoing debate in the analyst community about the advantages of software only innovation vs. hardware-software innovation (see Commodity hardware loses again and Commodity hardware always loses posts). Here is another example where two separate companies have turned to hardware innovation to take storage innovation to the next level.
DSSD D5 and FlashBlade
Within the last couple of weeks, two radically different AFAs were introduced. One by perennial heavyweight EMC with their new DSSD D5 rack scale flash system and the other by relatively new comer Pure Storage with their new FlashBlade storage system.
These two arrays seem to be going after opposite ends of the storage market: the 5U DSSD D5 is going after both structured and unstructured data that needs ultra high speed IO access (<100µsec) times and the 4U FlashBlade going after more general purpose unstructured data. And yet the two have have many similarities at least superficially.
Each of these systems is a new, ground up, hardware innovation, which uses raw NAND memory chips and provides all flash storage in blades/modules.
The EMC DSSD D5 uses a stateless, host client PCIe3 card which mimics a NVMe device for each server and connects via up to 96-PCIe Gen3 extended links to two D5 IO modules. The D5 has dual control modules that implement global flash management and data protection algorithms across all the NAND chips located on the flash modules. Each IO module can access any of the DSSD D5’s 36 flash modules to perform IO operations. D5 Flash Modules contain ASIC(s?) and NAND memory chips, has dual ported PCIe Gen 3 x4 connections and comes in 2 or 4TB storage configurations, in half or full populated modules, for an overall storage capacity of 36TB, 72TB or 144TB per system. Flash modules, control and IO modules, fans, and power are all hot pluggable and are designed to be customer serviceable. However, the client card is not hot pluggable. DSSD D5 provides 100 GB/sec of data throughput and millions of IOPS per system.
Each Pure Storage FlashBlade supports up to 15 blades which contains dual ARM core managed 8 or 52TB of superCap backed up, flash storage and an 8-core, Intel Xeon-based, scale-out storage controller, with dual-ethernet 10GbE connectors on the back connecting to the inter-cluster fabric. Each FlashBlade enclosure comes with two software defined ethernet networking switches and 8-40GbE connection ports, 4 per switch, to connect to other FlashBlades and hosts. The fabric interconnect uses a proprietary protocol across Ethernet. Hosts access the FlashBlade storage via the 40GbE fabric. FlashBlade blades, software defined networking switches, fans, and power are all hot pluggable and multiple FlashBlade enclosures can be connected together via the 40GbE fabric, to form a single scale-out storage system. The FlashBlade supplies up to 15GB/sec per 4U enclosure and a 2 FlashBlade chassis system can sustain up to 1M NFS ops/sec. FlashBlade storage starts at about 100TB and can go up from there.
There’s not many data management services supplied by the DSSD D5 and it does NOT provide data reduction but it does support thin provisioning, asynchronous and multi-threaded IO activity and atomic IO operations. The DSSD D5 also supports a two dimensional, Cubic RAID solution for flash data protection. DSSD control modules implement a global flash management layer which supports wear leveling, garbage collection, defect management, flash address translation, etc. for all the NAND chips. EMC mentioned that the standard OS IO stack takes anywhere from 300µsec to 1ms to get from an application’s request to PCIe NAND storage and back. DSSD replaces this IO stack with new, faster performing IO drivers so they can reduce this overhead substantially (100µsec). As such, customers have a choice of implementing a fully POSIX compliant, native DSSD block driver, a Flood Direct Memory API IO or DSSD provided plugins for specific middleware to deliver Flood Direct IO like performance. DSSD has already been working on an HDFS plugin to support D5 IO. The DSSD D5 is intended for structured or unstructured data using legacy or new applications which need extreme performance, such as Oracle, SAS, Real time analytics, etc. No indication on pricing for DSSD D5 storage, but it’s not going to be cheap.
Pure’s FlashBlade provides flash management, (3:1) data reduction, data encryption at rest and optimized meta data look up to support billions of files/objects and 10s of PB of file/object storage. FlashBlade storage is also optimized for any IO sizes. The management UI for the FlashBlade uses the current Pure Storage FlashArray UI, so current customers should feel right at home. Although FlashBlade only supports NFS and AWS S3 object storage today, plans are to support SMB/CIFS and HDFS, post GA. In addition, replication and snapshots are scheduled for a future release. With FlashBlade 3:1 data reduction, a 15-52TB FlashBlade chassis can provide up to 1.6PB of effective capacity. FlashBlade storage can be had for less than $1/effective GB.
Well there you have it. It’s seldom we see two storage systems with new hardware architectures going after opposite ends of the market come out within weeks of one another like this. They came out so close together and had technological aspects about them that were so similar, yet very distinct, that it’s hard not to draw some comparisons.
9 thoughts on “A tale of two AFAs: EMC DSSD D5 & Pure Storage FlashBlade”
A tale of two AFAs: EMC DSSD D5 & Pure Storage FlashBlade https://t.co/AnxSTXQlvI https://t.co/PH5QlTIafw
@RayLucchesi DSSD is not an AFA. It’s not an array at all. It’s essentially “pseudo-DAS” although it’s NVMe fabric, it acts like DAS.
@RayLucchesi actually neither product is AFA. Weird title.
Thanks for the comments Big J. As I tweeted, I would disagree, although one is file/object only and the other is NVMe both systems supply data storage to multiple hosts and are using all flash storage as their media.
RT @RayLucchesi: [Blog post] A tale of two AFAs: EMC DSSD D5 & Pure Storage FlashBlade https://t.co/II8sr4eSrj
Feedbin star: A tale of two AFAs: EMC DSSD D5 & Pure Storage FlashBlade https://t.co/CR727nTyyF
#SymLink: A tale of two AFAs: EMC DSSD D5 & Pure Storage FlashBlade https://t.co/8bsaLpWeXP
The two systems you mention – at the extremes of the storage performance vs capacity – illustrate that convential spinning disk storage – has been overtaken in performance, capacity and TCO by flash memory modules. It would be worth a mention of software like EMC ScaleIO that allows the new 4TB and 12TB solid state disks to be used in servers as a shared storage pool SAN, while convential Storage arrays – also all flash – have become both storage and data services platforms.
Thanks for the comment.
ScaleIO and the other Software Defined Storage solutions are taking SSDs to new heights in performance, usability and in some cases, economics. How this goes on long term and what it means for disk is another question entirely. Both of these systems support some form of shared storage, although the DSSD D5 may be more of a carve it up and only one dual ported host can access it kind of “shared” storage (don’t really know for sure). FlashBlade is an NFS solution so it’s inherently shareable…
ScaleIO, VSAN and a few others are leading the charge from the Software Defined end of things, keep up the good work.
Comments are closed.