Storage lagging behind servers
But there is a looming problem here and Darren Thomas, Dell's storage VP and general manager, talked about it yesterday. He said that today's PCIe bus is about as fast as Fibre Channel, meaning around 8Gbit/s, and Intel's next-generation processor memory bus will be 100 times faster - this was a big picture pitch. What that means, he says, is that servers will be waiting for networked storage and that won't do. He says he thinks he knows how to fix that but isn't ready to talk about it.
The answer would seem to be binary. Either the storage gets an order of magnitude or two increase in network link speed to the servers, or NAS and SAN give way to DAS (directly-attached storage) and are directly connected to servers somehow. The storage data gets to servers much, much faster, or the storage becomes integrated into the servers.
The net of this is that Dell's storage has to get used to faster servers wanting data fed into their memory faster than ever. Flash will be provided as a fast storage tier adjacent to the server's DRAM, and that flash has to be loaded from disk-based resources that can cope with the bandwidth demands made on it. That means the storage array software has to change.
Dell's storage has to get used to faster servers wanting data fed into their memory faster than ever.
Intel says that current second-generation PCIe does 5 giga-transfers/second (Gtf), meaning up to 20Gbit/s. Gen 3, coming with Romley processors in the first quarter next year, will run at 8Gtf and could do up to 100Gbit/s.
Storage's ambit will extend to tier zero server flash which will have, we imagine, a PCIe gen 3 link. So the bulk of the storage resource, tiers 1, 2 and whatever else, will need to feed this flash and accept writes from it. It could do it by having its own PCIe gen 3 interface or going in via an Ethernet port, presumably a 10Gbit one, meaning 10Gbit between the network interface and the storage. Arguably this would be too slow and a 40Gbit/s NIC would be needed.
Dell storage people aren't talking about any of this publicly. But it seems clear that the Compellent and EqualLogic controller software will have to manage tier zero flash and the externally-attached array becomes an internally-attached array either directly PCIe gen 3-connected or coming in via a NIC.
An interesting thought: if both EqualLogic and Compellent arrays become internally-attached and do server IO through a PCIe gen 3 interface then differentiation based on iSCSI vs Fibre Channel access falls away. It's all DAS - unless the PCIe bus carries a storage networking protocol.
Dell and innovation
What we are seeing here is a server-storage technology gap that Dell is filling from within its own resources. The lines between servers, networking and EqualLogic and Compellent storage are blurring with lots of staff meetings held to ensure that developments in each area are in lockstep.
Dell is now definitely focussing on innovation. This is no way and no longer a me-too technology company. Dell is upping its capabilities and taking on anybody and everybody in an IT world where separate server, network and storage boxes and silos are going away.
Big changes are coming and Michael Dell, the boss of the biggest IT company in Texas, its lone IT star we might say, is putting on his ten gallon hat and doing what an IT man has gotta do to make his way in the big data-driven cloud computing world he sees coming. The man from Texas is starting to swagger because he reckons he's got the answer. ®