In a recent blog, EMC Corp. vice president of technology alliances Chuck Hollis made a case for NAS (network-attached...
storage) rather than a SAN (storage area network) in large VMware Virtual Infrastructure 3 (VI3) environments. EMC owns VMware, but runs it as an independent company.
Today, the vast majority of VMware ESX hosts are attached to SAN storage – at least 70%, said Hollis.
Others, like Rod Lucero, CTO at the VMware reseller VMPowered in Minneapolis, Minn., put that number closer to 90%.
NAS has numerous advantages over SAN in a VMware environment, said Hollis in a follow-up interview. First and foremost, "a NAS namespace is a lot easier to manage than a SAN namespace," he said.
That's especially true in large VMware environments that make heavy use of VMware's VMotion and Distributed Resource Scheduler (DRS), and both features are available with VI3. DRS works with VMotion to automatically migrate virtual machines (VMs) between ESX hosts based on performance and availability policies.
The problem with DRS in a SAN environment is that, in order to make effective use of it, "you're going to need to open up your SAN to all your servers," Hollis said. From a best practices perspective, that can be "potentially problematic," Hollis said.
Those concerns can be mitigated by using SAN techniques such as zoning and LUN masking, but a much simpler alternative might be to simply use a NAS file system as the data store for VM images.
"I think that, in the long term, we'll find high-end NAS much more friendly for high-end VMotion / [Distributed Resource Scheduler] farms than today's SANs," Hollis blogged.
As an aside, Hollis said he has seen a dramatic increase in the size and sophistication of VMotion and DRS deployments. "At first VMotion was used mainly for server-to-server failover; that was the state of the art in 2005 and 2006 timeframe," he said. The introduction of DRS last summer as part of the VI3 release, however, laid the groundwork for "the optimization of service levels as opposed to simple hardware failures." Hollis personally knows of about two dozen EMC customers doing large-scale load-balancing with DRS.
File over block
Management isn't the only benefit of using NAS in a virtualization environment, Hollis said. "NAS has the potential to offer a few benefits that we might not find in the SAN world," Hollis wrote.
For one thing, NAS relies on ubiquitous and inexpensive TCP/IP and Ethernet cabling, not Fibre Channel. It also provides for decent security and access control via TCP/IP and NFS.
Furthermore, interacting with a file system makes it easier to do tiering -- positioning data on different classes of storage based on its performance and access characteristics. "Tiering can be done in block, but if you have a file presentation, you gain access to a wealth of [information lifecyle management] tools," he said.
VMPowered's Lucero, meanwhile, recommends NAS over SAN storage to his customers in cases where they'd like to replicate their storage. With a SAN, if you want to do replication, you have to replicate the entire LUN, Lucero said. With NAS, you can use file-based replication and only copy the changes to the individual files, i.e., virtual machines.
Lucero started using NAS to store VMs seven years ago at a previous employer "because that's what we had available." When VMware introduced ESX 1, the company discontinued support for NAS, but it supports it again as of ESX 3.
Where NAS nosedives
If NAS is so great, why doesn't everyone use it?
In a nutshell, for performance reasons, said Rob Stevenson, managing director for storage at TheInfoPro Inc., a market research firm focused on the IT operations at Fortune 1000 companies. With NAS, "the TCP/IP stack has to be accessed every time you access your data, so you don't get nearly as good performance," he said.
But according to Hollis, the NAS performance penalty may be overstated. "To be honest, we're not seeing a whole lot of high performance stuff being put on VMware," he said. Furthermore, whether a packet is traveling over Fibre Channel or Ethernet, "it still travels at the speed of light; the real issue is bandwidth – do I need more lanes on the highway?" High-end NAS platforms such as those from EMC, BlueArc Corp. and Network Appliance Inc. can be configured to transmit data over multiple gigabit Ethernet ports, matching the bandwidth of a single 4Gbps Fibre Channel pipe.
Stevenson said that in order to see if NAS is a good fit for an environment, simply add up the aggregate I/Os per second that you expect the ESX hosts to generate.
For Lucero, that usually translates into placing virtual machines running "non-transactional" applications on NAS -- for example, Web servers, DNS and DHCP, or common network services like Cisco fabric management apps.
But that will change, Lucero said. "Over time, NAS is going to become much more of a player, especially as we get to 10 Gig Ethernet."
Let us know what you think about the story; e-mail: Alex Barrett, News Director