Replication Concerns Carry Into Virtualization

News from at least one vendor this week begs the question: Is it easier to replicate in virtual environments than in physical ones?

DataCore Software announced yesterday that Tokyo-based SoftBank Telecom used its SANmelody software in a proof-of-concept (POC) project featuring remote-site mirroring of a Tokyo data center in Osaka, Japan, about 500 kilometers away. The setup is part of a planned design to support new services from SoftBank, DataCore says.

In the POC implementation, DataCore's software resides on a dedicated server in an InfiniBand-based network (based on Bay Micro switches). Another server in the network runs VMware ESX and VMware's ESX HA (high availability) software. Virtual machines generated by the VMware hypervisor use the DataCore server to front-end about 32 Tbytes of Fibre Channel storage (from IBM's DS Series). The DataCore server creates shared storage and in turn uses multiple "virtual iSCSI HBAs" created by a connected appliance from Xsigo as the networking mechanism for mirroring data between sites. The Xsigo appliance works with the router to send the data over the WAN via Sonet.

"It is clear that the combination of virtual storage and virtual servers in an overall virtual infrastructure make it much easier to do disaster recovery," stated Peter Thompson, VP of Asia Pacific for DataCore.

But industry sources say storage managers face the same issues of performance, manageability, and vendor lock-in, whether they're replicating data from virtual machines or "real" ones."I don't think that data being virtual really changes anything," says analyst W. Curtis Preston of GlassHouse Technologies. He says users still face the choice to replicate data using storage from array vendors, or software-only solutions.

Conventional views hold that storage arrays featuring replication, such as those from BlueArc, EMC, NetApp, and Reldata (to name a very few), offer the advantage of hardware-based performance. Many of these suppliers have certified their wares for VMware and other virtual infrastructures, so they can tackle a range of storage functions for data conveyed in virtual machines or physical ones.

In contrast, software-only products, such as those from DataCore, LeftHand Networks, and others, can be cheaper, the thinking goes, and will support multivendor arrays and often more flexible management functions.

"With array-based replication, you need to have the same array on the remote site," says Arun Taneja of the Taneja Group consultancy. "That can be very expensive." Solutions that virtualize storage like DataCore's or LeftHand's, he says, allow users to replicate data (from virtual or physical servers) from one vendor's array to another's, if they want to.

DataCore isn't elaborating on whether the arrays at both the SoftBank sites are from different vendors, and SoftBank isn't available to comment. But the POC solution suggested by DataCore could conceivably take place with a range of other solutions."You can replicate in hardware- or software-based arrays; you can do it in the network as a fabric service; you can do it in the server or in the guest of the virtual environment," says Greg Schulz of the StorageIO Group.

Ultimately, every shop will have to choose based on their own criteria, Schulz notes. Some may find it costs more to add storage virtualization software like DataCore's to their networks than to buy another array. Others may view the investment in remote-site arrays as justified by performance.

While there may be little distinction between the issues of virtualized versus physical replication, it's clear that virtualization is forcing some customers to look more closely at how they're working storage.

One manager, Jeremy Moss, group IT director for Anite, an IT services and integration firm in the U.K., is perhaps typical. Moss and his group moved to VMware Infrastructure three months ago, realizing they needed the data resiliency and failover options that environment offered. But implementing VMware required shared storage, and up to then, Anite had relied on DAS.

Moss says his group picked LeftHand Networks's software because it was cheaper than a NAS and wasn't tied to a particular hardware solution. "We are keen on hardware independence," he noted. His team evaluated a NetApp system, but he says it cost "twice as much" as the LeftHand software, which had the added benefit of running on Intel servers. "Once we made the choice, we also liked that we could do intersite replication using the same [LeftHand] software."Have a comment on this story? Please click "Discuss" below. If you'd like to contact Byte and Switch's editors directly, send us a message.