In-Depth
Of Knee-Nippers and Ankle-Biters
It should come as no surprise that new companies have begun to carve out a part of the storage market for themselves.
A fiscal quarter or two ago, when EMC and Network Appliance, both big players in the storage arena, reported flat or negative earnings from sales of their arrays, one Wall Street analyst attributed their rwaeversal of fortune to the increasing sales of the “ankle biters” appearing in the industry. The term stuck with me because I have children ranging in ages from 21 to 5. I know something about ankle biters (toddlers) and about “knee nippers” (somewhat taller sub-teens).
Fleshing out the “ankle biter” comment, the analyst pointed specifically to LeftHand Networks, 3PAR, Compellent, Equallogic, Adaptec, and a few others, stating that their lower prices for comparable functionality was allowing them to take some market share from their older competitors.
If I had to vote, I would classify the companies mentioned as “knee nippers” rather than ankle biters. Most have been around since the late 1990s and have been steadily building channel organizations to push their products. Several were pioneering iSCSI storage before it was popular and are now coming into their own as iSCSI use is on the rise. Others have been vociferous in their advocacy of new/old ideas, such as Thin Provisioning, that theoretically enables you to allocate for other uses that portion of storage capacity that is reserved but not yet used by applications—in a kind of storage shell game. (So popular is Thin Provisioning, EMC is now promising to deliver it on arrays in 2008.)
These firms are clearly no longer in diapers. They have achieved knee nipper presence. In other words, they have achieved a combination of stature and assertiveness that allows them to collide, inevitably and painfully, with the solar plexus of any adult in the room.
Clearly, companies such as LeftHand and Adaptec have achieved such stature and assertiveness. LeftHand has been racking up sales for several years, and their balance sheet was in the green the last time I checked. Their head clustering approach, which we are testing in our labs, is innovative and blows the socks off of the NetApp products that leverage the ideas of Spinnaker Networks, acquired by Sunnyvale a few years back.
As for Adaptec, Steve Rogers and company who oversee the SNAP product family have consistently put together some of the highest performance NAS and iSCSI arrays that we have tested or seen in the market. With their latest wares, to be announced in September, you will get just about every bell and whistle that you see in a NetApp filer for a fraction of the price.
It comes as no surprise that these companies have begun to carve out a part of the market for themselves. It’s that whole circle of life thing.
When I think of ankle biters, I think of a younger generation of storage companies that includes JMR Electronics, Coraid, and Zetera. JMR is no newbie to storage, to be sure. Founded in 1982, the company was one of the first producers of RAID systems fifteen years ago and has been a provider of customized storage systems ever since.
Expanding the PCIextended Bus
A few weeks ago, JMR founder Josef Rabinovitz sent me a copy of a book he had written and self-published with the unassuming name, The Next Generation Storage. It described a technology for expanding the PCIextended bus architecture, increasingly common in servers and desktop systems, to create a new kind of direct-attached storage, which he calls PeSAN.
Think mainframe bus and tag channel attachment, if you have ever played in that space. Rabinovitz argues that simply by not converting the PCI native bus into any other expansion bus architecture (SCSI, Fibre Channel, et al), PeSAN delivers all of the performance of other storage approaches, but without the complexity, cost, and other possible points of failure.
With PeSAN, you can expand the storage using switches that speak PCIe. Plus, you can gang switches together with very little throughput degradation, according to Rabinovitz, enabling the creation of large complexes of SAS and SATA storage arrays. Remarkably, the PeSAN architecture comes in at about fifty cents per GB for an 8TB JBOD (just a bunch of disks, non-RAID array), versus $2.88 for a comparable SCSI JBOD, or $3.64 per GB for a FC JBOD of the same capacity.
Jim Kemp is also a proud ankle-biter, though he might bristle at the terminology. Since 2000, his Athens, GA-based company, Coraid, has been creating a storage technology that leverages Ethernet itself as the mechanism for transporting reads and writes to ATA-compatible disk. His protocol, ATA over Ethernet or AoE, has been deployed by over 1000 customers, making the privately held company profitable for the last seven consecutive quarters.
AoE has been a darling of the Linux community, which embraced his protocol stack early on and embedded it in some of the more prominent kernel builds. Kemp has seen uptake in three main categories of customers: system integrators who use AoE in their solutions, hosting companies and ISPs who find it to be a quick and less-expensive route to building massive repositories for data (one customer has deployed over a petabyte of Coraid storage to support voice recognition services), and the government and university community, which has embraced Linux in a big way and prefers the low cost, native attachment method from Coraid that leverages ubiquitous standards like Ethernet and ATA.
Coraid’s next big announcement is a virtualization appliance that will ship at the end of August to address some of its customers’ needs for allocating storage at the dynamic LUN level. According to Kemp, many of his customers have used LUN management utilities like LVM2 in the Linux OS, which “plays much the same function as Symantec/Veritas Volume Manager in other OS environments.” However, in mixed OS settings, customers wanted another approach.
“Some of our customer shops are pure Linux,” Kemp said, “and they were happy with the ability of our storage appliances to RAID and share. However, others may be using, in addition to Linux, Microsoft for e-mail, or Apple for video work. They wanted to virtualize storage and provide LUNs that could be accessed by all environments.”
The EtherDrive VirtualStorage Appliance delivers the goods in a novel way, according to Kemp. It performs on-the-fly LBA address translation, so reads come directly from the storage and not from the virtualization appliance. As writes are made to storage, frames are readdressed to the appropriate targets. However no data is cached, which in Kemp’s words “allows you to build a clustered storage solution with heartbeats in a way that is easy to implement and scale.” A datasheet on the product is available from Coraid’s Web site.
Over in Irvine, CA, Zetera Corporation continues to drive its ZSAN story—storage connected via UDP and IP. As previously covered in this column, ZSAN drives cost out of storage by leveraging the multicasting functionality of IP enabled by using UDP as a transport mechanism. With Zetera, you can connect up an almost limitless number of storage disks, assign them IP addresses automatically with DHCP, then group them into broadcast/multicast subscriber groups to create virtual volumes that can scale dynamically. The resulting volumes show up as drive letters on any system that you want to have access to them and they perform like locally-attached disk.
Virtualization at the protocol layer is pure elegance—and driving cost out of storage by eliminating the need for controllers, specialty switches, RAID, and value-add virtualization software or hardware has huge appeal to companies large and small. As proof, consult the statistics cited by CEO and co-founder, Chuck Cortright.
Since the introduction of Zetera’s technology in 2002, which (like the aforementioned companies) is steeped in patents, five OEMs, including U.S. powerhouses NetGear and Bell Micro, have begun offering products based on ZSAN in every corner of the globe. Over 200,000 units have shipped, including a multi-petabyte solution deployed at MIT’s Media Labs.
Arguably, ZSANs have outsold FC SANs in their first couple of years and buy-in continues from stalwarts in the industry. Stand by for news of some significant investors in this company over the next couple of months.
So there you have it. The big-iron guys need to watch their solar plexus as the knee nippers provide comparable functionality at better prices. Coming on the heels of this wave of competitors are the ankle biters who aspire less to be copies of their older kin than innovators with a new concept of storage altogether. If I were EMC, IBM or HDS, I would think about buying these guys—if only to keep them in their cradles.
Your opinions are welcome: [email protected]
.