EMC's un-announced virtualising Storage Router was demonstrated on the Cisco booth at CEBIT. It involves routing tables running on a module in the switch plus EMC software running in an attached appliance. This isn't routing in the sense that MCDATA might use the term, to signify routing between two separate SANs to unify them but keep faults isolated to just one. EMC is seeing storage virtualisation as a layer 3 networking function, like routing. THe actual product name should be different.
Mark Lewis, executive VP for EMC Software, says: "It will be formally launched at ETS, EMC's conference in New Orleans in May. It's been in beta since October in nine large customers."
What's it for?
"It's goal is to create a thin layer of abstraction between storage arrays and applications that use them. It presents a virtual volume.The sort of thing it would do is to migrate data from one storage array to another."
Would the data copy be done and the virtual volume map changed once the copy is complete?
"Data is copied in chunks and the mapping for a transferred chunk is changed." So you use both arrays as the transfer is carried out. There is no loss of service.
Another use would be to do this, then upgrade the old array with new firmware, controllers, whatever, then bring it online with no loss of service whilst the upgrade has been carried out.
"You can span data across arrays. Ultimately we can build much better storage services with different levels: gold; silver; and bronze for example. We can create continuous data availability for customers. They could move a data centre and restrict access to the data whilst doing it."
Lewis didn't mention getting rid of wasted space on disk arrays but no doubt the EMC virtualisation facility will let you do that too.
For EMC: "Virtualisation belongs everywhere: In the RAID controller and disk arrays (Symmetric and CLARiiON); in the server (with VMWare); in the storage network; and in the SAN with, for example, Cisco's VLAN. With virtualisation everything will run better.
What is the hardware?
"We have partnered with the switch vendors. They create the ASICs. We put the software, the routing tables, into their hardware. It's running at hardware speed to do the routing. There's very low latency."
EMC writes its Storage Router software to a standard, FAIS for Fabric Application Interface Standard. Lewis says: "It will be adopted by Brocade, Cisco and McDATA. They've all agreed. They'll compete on their ability to build products behind it. We get to write to a single switch interface."
"We could run with one switch from Cisco and one from Brocade." FAIS is that good.
"We don't even run any software in the switch. We just load routing tables into the switch. Our software will run as an appliance." This is how EMC gets the advantages of being out-of-band for control software but in-band for data transfer. This technique was called SPAID (Split Path Architecture for Intelligent Devices), a term coined, according to Lewis, by an analyst wanting to give EMC's product some distinctiveness. It isn't an EMC term.
Cisco calls the EMC software an intelligent fabric application and it runs on a storage service module. This is a specialized 32-port Fibre Channel line card that can operate on the Cisco MDS 9500 Series directors and the Cisco MDS 9200 Series fabric switches. It requires SAN-OS 2.1, the latest version of the MDS 9000 O/S, which includes features designed specifically to work with third-party storage software.
The storage appliance running the EMC software is just an Intel box. It's nothing special, although it is fault-tolerant. The software will be developed: "We're going to have appliances out there to provide services such as replication, etc."
Cisco needs third party applications to use its hardware and is getting them from EMC, IBM and Veritas. Todd Oseth, EMC's VP for Storage Infrastructure Software, said: "EMC and Cisco share a common view that, in order to maximize performance and scalability while also protecting customers' existing investments in functionality, storage virtualization is best implemented on the storage network." Bob Mahoney, a business line executive at IBM, said: "IBM and Cisco have been partnering for years in high-performance, scalable storage virtualization technologies including Cisco Virtual SANs and IBM's SAN Volume Controller. These new offerings will enable even more integration of IBM's business continuity capabilities into the intelligent fabric."
We're going to see all the switch vendors offering fabric platforms for running storage service applications from EMC, including Documentum and Legato, IBM, Veritas and, no doubt, many others.
How does EMC's virtualisation differ from the competition?
Lewis says that competing virtualisation vendors cache the data: "We don't do that." Who caches the data? "Both IBM with its SVC and HDS with its TagmaStore do it. Both are trying to virtualise data by, in effect, putting an array controller in the network because they don't have the hardware we do."
(But this, for IBM, seems plain wrong as IBM's SAN Volume Controller also runs in the Cisco MDS 9000.)
Lewis says this caching is a problem: "Data isn't considered stored until it's in the array. If the network has to guarantee storage integrity then it will cost them major problems."
Another issue is heterogeneous drive array support: "TagmaStore has to work with all of its supported arrays and do a tightly-coupled qualification. Storage Router will support all the arrays in the network and all the server operating systems. It's inherent." Virtualisation across heterogeneous vendor arrays in a SAN is taken for granted with storage router and, since its out-of-band, there is no dependency on the accessing server hardware/operating system platforms'
Yet another point of difference is replication: "Take heterogeneous replication; our Storage Router can do that. With HDS' TagmaStore you can't do synchronous replication since you cache it in the network. All relevant business policies are destroyed."
HDS would probably beg to differ. A spokesperson said: "Hitachi handles the virtualisation inside the USP (TagmaStore) using TrueCopy Asynchronous and Hitachi’s Universal Replicator to replicate both internal disk and virtualised external disk storage without the need for any additional appliances."
A final point of difference identified by Lewis is disruption: "IBM has been selling SVC for three years and not done a lot with it. It and TagmaStore are very disruptive to the existing environment. We'll go in (with Storage Router) without disrupting the business processes. It's the same as with VMWare. That's very important to the adoption of innovations."
IBM would almost certainly beg to differ here. It has around 1000 customers using its SAN Volume Controller, with 80 percent or more in full production and has been busy establishing the market that EMC is about to enter. Some customers are listed here. IBM began shipping the SVC appliance in July 2003. So it's been selling an average of over fifty a month. The disruption point Lewis mentioned might not actually count for much.
Note. HDS has just upgraded TagmaStore. It has added multi-vendor virtualisation facilities so that almost all drive arrays from EMC, HP and IBM, and of course HDS itself, can be virtualised behind TagmaStore. HDS says both synchronous and asynchronous replication can be carried out between these different arrays.