The Portland Group
CSCS Top Right Frontpage

Since 1986 - Covering the Fastest Computers
in the World and the People Who Run Them

Language Flags

Visit additional Tabor Communication Publications

Enterprise Tech
HPCwire Japan

Blog: From the Editor

From the Editor | Main Blog Index

High Performance Storage Getting More Flashy

The penetration of solid state storage into high-end computing seems to have picked up recently. Just this week we saw a couple of developments that may signal flash technology has got its second wind.

On Monday, Sun Microsystems finally announced its much-anticipated F5100 Flash Array, a 1U storage appliance that can house up to 2 TB solid state storage. Sun is claiming a single F5100 can deliver an impressive 1.6 million read and 1.2 million write IOPS, while consuming just north of 300 watts. Performance-wise, Sun says, that is equivalent to 3,000 hard disk drives,which will take up more than 14 datacenter racks and consume more than 40,000 watts. Of course, that many disk drives are going to provide petabytes of capacity rather than terabytes, but if IOPS are your thing, solid state is the obvious choice. It doesn't come cheap, though. Sun quotes a list prices of $159,995.00 for the 2TB configuration.

Given the impending Sun-Oracle marriage, the F5100 is being billed as a storage accelerator for Oracle and MySQL database workloads. But the storage array can also be recruited for HPC duty. For example, the San Diego Supercomputer Center (SDSC) is already kicking the tires on the F5100. From the Sun press release, Don Thorp, Operations Manager at SDSC had this to say:

"San Diego Supercomputer Center (SDSC) has been evaluating the F5100 Flash Storage array as a high performance SamQFS metadata target, which sits at the core of our archiving services and hosts well over one hundred million files. Performance improvement of 2.5 to four times was demonstrated for file creation and metadata scans, such as listing and backups. Further testing will be done using the Sun Storage F5100 as a Lustre metadata target, high speed storage pool in Lustre 2.0 for user checkpoint data, Oracle database storage device and out-of-core storage device on an HPC cluster."

SDSC, by the way, seems to be on a flash storage binge lately. In September, the center announced "Dash," a new Appro super accelerated by Intel SSDs. I'm guessing most of the other supercomputer centers and DOE labs will be joining the flash-fest over the next year or so.

The second bit of news this week comes from Fusion-io. Apparently social networking site MySpace has gone gaga for Fusion's ioDrive. From the press release:

The MySpace installation consists of replacing multiple server farms of larger two rack unit high (2u) servers that require 10-to-12 15,000 RPM mechanical disk drives each with much smaller 1u servers that use Fusion-io's server-deployed ioDrives. In addition, ioDrive-deployed servers can now replace MySpace servers that hold everything entirely in RAM – a cost prohibitive and power hungry approach previously required to achieve necessary data throughput now attainable through the high-performance, solid-state solutions from Fusion-io.

The case study (PDF) referenced in the press release said by using the flash devices, MySpace was able to replace150 of their standard load servers and 50 to 80 heavy load servers, allowing the company to save 215U of rack space. The study went on to say that because the flash replacement provided substantial power savings per server (50 percent), the reduction in costs should "easily pay for the ioDrives over their lifetime."

It's worth noting that neither the new Sun product nor the Fusion ioDrive are using flash in the form of solid state drives. In the case of the Sun F5100, flash modules are incorporated into a standalone storage appliance with an SAS interface. The ioDrive solution connects flash modules to a PCIe bus, thus avoiding the disk controller bottleneck and allowing the fast storage to get much closer to the CPU.

Other vendors are starting to offer PCIe-based flash for servers as well, either as standalone cards, such as Texas Memory Systems' RamSan-10/20 offerings, or incorporated into appliances like Dolphin's StorExpress. Sun offers its own PCIe flash card, the F20, which shares some of the same technology found in the F5100.

Now that solid state storage vendors are starting to come up with more mature products, and early adopters are starting to sing the praises of flash, I would expect to see more action in the HPC space over the coming months. I would also expect to see some of the more creative OEMs and system integrators to start building this technology into HPC cluster and storage products in the near future.

Posted by Michael Feldman - October 14, 2009 @ 4:25 PM, Pacific Daylight Time

Michael Feldman

Michael Feldman

Michael Feldman is the editor of HPCwire.

More Michael Feldman

Recent Comments

No Recent Blog Comments

Sponsored Whitepapers

Breaking I/O Bottlenecks

10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.

A New Ultra-Dense Hyper-Scale x86 Server Design

10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.

Sponsored Multimedia

Xyratex, presents ClusterStor at the Vendor Showdown at ISC13

Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.

HPCwire Live! Atlanta's Big Data Kick Off Week Meets HPC

Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?