In-Depth

Pick a LUN, Any LUN

Familiar problems are starting to crop up

• By Jon William Toigo
• 05/20/2004

For a time, we were all told by research firm Gartner and others that there were four kinds of virtualization: one based on host software, one based on an in-band appliance or switch, another based on an out-of-band appliance, and a fourth kind based on array controllers themselves. This convenient categorization of virtualization schemes served the interests of the analysts because they could keep their vendor clients out of each other’s “magic quadrangles.” Two solutions would not be compared with each other because they were segregated into entirely different categories. The industry likes this because they hate for analysts to actually compare any products to one another.

The long and the short of it is, the differentiation was mostly specious from a technical perspective. All virtualization is in-band – I/O must pass through the virtualization layer that is created in order to get itself placed onto storage LUNs aggregated by virtualization software into a volume. At the end of the day, it was a matter of how you wanted to license your software and how much you wanted to pay. Host-based approaches required every server to have a software license. Controller-based approaches contributed to the high cost of storage arrays. Everything else was a one-stop-shop purchase, sold by the pound of storage under management or the number of application servers communicating across the in- or out-of-band server appliances.

One thing that the analyst idiocy did create was a holy war over in-band and out-of-band approaches that was fueled in dueling whitepapers and marketing messages between DataCore Software (an in-band appliance vendor) and StoreAge Networks (an out-of-band appliance vendor). Successful implementations of each technology should have put the arguments over solution efficacy and potential chokepoints to rest. However, you can still hear folks ruminating about the “intuitively obvious” chokepoint created using an in-band appliance. Intuitively obvious or not, there is no chokepoint when you apply healthy dollops of fast processing, parallelism, and low interrupt I/O pathing to the in-band appliance, as DataCore has proven.

Microsoft hurt vendors of the host-based schemes, in Windows shops at least, by adding volume management capabilities directly into Windows Server 2003. Almost every vendor in this space saw the handwriting on the wall and started looking to make nice with switch makers in 2002 in order to find a new deployment platform for their products. And the array guys continued to do what array guys do: charging a lot of dough for microcode added to their boxes of Seagate disk drives.

Bottom line: By late last year, it looked like the virtualization story was finally getting sane. Until recently, that is.

IBM announced yet another in-the-wire appliance, its SAN Volume Controller, touting its ability to aggregate LUNs from any vendor into a common storage pool for use in building volumes and expanding volume capacity on the fly. It looked like the concerns of many people (including me) about the resistance of highly competitive vendors (such as EMC, IBM, and Hitachi) to the concept of sharing out their storage into a commodity pool were finally about to be laid to rest. Even John Tyrrell, until very recently a senior architect with EMC, said that differences between LUNs were gone and that the only obstacles to combining LUNs from different vendors were political, not technical.

Then, CNET published an article quoting an unnamed source at EMC to the effect that IBM’s SAN Volume Controller could NOT be used with EMC gear—Symmetrix or Clariion. According to the source, EMC LUNs were a breed apart from other vendor’s LUNs and combining them indiscriminately with lesser LUNs would pollute, confuse, and otherwise create havoc in your EMC storage. At about the same time that the story was going to press at CNET, EMC CEO Joe Tucci was on stage in Orlando talking up a storage virtualization router that Hopkinton was planning for 2005.

So, just when we thought that the whole virtualization cold war was subsiding, and an Age of Aquarius in LUN aggregation was about to begin, it looks like we need to break out the cold weather gear once again. So it goes in the storage world.

Let me know your thoughts. jtoigo@intnet.net