At the recent EMC Forum in Long Beach, Calif., more than 800 customers gathered to hear the inside scoop on where the company is heading. It was a day dominated by highly technical sessions. Even the keynotes were high on content and low on PR.
"We are building the equivalent of a mainframe for open systems," says Michael La Fauci, technical business consultant for EMC's emerging technology practice. "We are talking about mainframe-like functionality, but with decentralized CPUs, memory and intelligence and centralized management."
You Name It, EMC Can Virtualize It
EMC's virtualization message spans multiple facets. David DeWalt, the company's president of worldwide customer operations (and formerly the head of Documentum), laid out the three main elements:
- Server virtualization via VMware;
- File level virtualization via Rainfinity; and
- Block-level virtualization via Invista.
One forum attendee highlighted the fact that he has reached the point where he can consolidate 30 white boxes onto one larger server using VMware. This very much fits into the general understanding of what virtualization is.
"Virtualization provides logical views of physical resources while preserving their usage interfaces," says Tom Bailey, an EMC virtualization specialist. "It removes physical limits and improves utilization."
While VMware addresses servers, Rainfinity is aimed at file virtualization of multiple NAS boxes. IDC reports that 65 percent of enterprises are planning to implement file virtualization in the next 18 months. Other analysts report similar trends.
"NAS virtualization is one of the key developing market segments right now, with lots of competing products and a great deal of innovation," says Stephen Foskett, director of strategy services at GlassHouse Technologies. "It ranks with CDP, archiving and iSCSI as the key product areas in storage in 2006. However, end users are just starting to get the message, and I expect it won't be until 2007 that we see purchases coming on strong."
File virtualization creates an abstraction layer between where files and directories are stored and where they are viewed. This allows the storage infrastructure to change as needed without disturbing clients and the applications that use files. Such features reduce the time it takes to manage file server tasks such as adding capacity, migrating content or performing maintenance. It also helps in consolidating unused capacity and rebalancing loads.
"If every NAS device is physically and logically independent, migrations are extremely complex," says Jack Norris, a file virtualization manager at EMC. "Virtualization makes it possible to transport a lot more data without any need to coordinate with end users."
He gives the example of an ISP with more than one million users. It increased utilization rates on its NAS boxes by 30 percent using Rainfinity. He also talks about arbitrary tiering of storage architectures, which can lead to Tier One becoming heavily over-utilized while Tier Two sits underutilized. Virtualization, he says, can add more flexibility to tiering in order to optimize what is stored where.
Rainfinity is sold as an appliance that generates synchronous file replication over IP. Instead of wrestling with permissions, domains and security issues for each NAS box, this technology reduces the complexity of a data move to a few clicks. It also operates beyond EMC NAS to virtualize other OEM gear. To accomplish its functions, it creates a private VLAN. For data movement over the WAN, an appliance is required at each end.
Block Virtualization Party
EMC techie La Fauci explained his view that virtualization represents something of an expanding universe. The SAN started it all, then NAS included file servers within its perimeter.
"The last part being virtualized is the overall heterogeneous storage infrastructure," he says. "With Invista and Kashya, we can reach just about anywhere in the enterprise."
He explains that there are five layers within the Fibre Channel protocol. Layer 3 is services and virtualization that's where Invista sits. It offers a layer of abstraction to block-level storage.
Invista software has a discrete box that connects over IP to an intelligent switch such as a Brocade SilkWorm AP7420 or Cisco SSM. The APIs of these machines (XPath and SANTap, respectively) do the heavy lifting.
"Invista is only doing about five percent of the work and that concerns error conditions," says La Fauci. "It is much better to distribute virtualization all across the I/O path than try to tie it all into one box."
Currently, Invista only runs on a Solaris host. By the end of the year, AIX and HP-UX will be added, followed by Microsoft in early 2007.