A new virtualization replication company called Zerto claims to have built a better mousetrap for disaster recovery.
Zerto left stealth mode last month and is already generating some buzz, because it doesn’t require matching storage arrays at disaster recovery sites and allows users to manage replication at the VM or application level, rather than at the physical storage layer. Its software installs as a virtual appliance on each host, using vSphere APIs and Zerto’s own proprietary algorithms to replicate virtual machine (VM) data to a secondary site.
By operating at the hypervisor level, Zerto aims to eliminate the complexity that can arise when replicating between storage area networks (SANs). Traditionally, users often have to match up VMs, VMDK files and their underlying physical volumes on the SAN during replication and recovery. But with replication at the hypervisor level, according to Zerto’s data sheet, users can select which data to replicate according to which VM they want to protect, or they can specify which VMs should be grouped together for replication as a multi-tiered app. Data still needs a place to be stored at the secondary site, eventually, but because matching arrays aren’t needed, the storage at the secondary site can be cheaper than what’s used at the primary site.
Zerto’s virtual appliances, called Virtual Replication Appliances, communicate with one another, so if VMs or their VMDK files are moved — through vMotion, High Availability, Distributed Resource Scheduler or Storage vMotion — replication remains consistent. The product compresses data to cut down on network traffic over the WAN, integrates with vCenter for management and reports on replication operations.
Woodforest National Bank, Zerto’s current poster-child customer, is about 96% virtualized, running about 300 VMware vSphere 4.1 ESXi virtual machines on about 25 hosts at each of two sites in The Woodlands and Austin, Texas. The bank is testing Zerto’s replication for disaster recovery for all of its servers, CTO Richard Ferrara said.
Currently, the bank is using EMC’s RecoverPoint, based on Kashya’s replication technology, which was sold to EMC by the founders of Zerto in 2006. The bank’s RecoverPoint configuration uses the SANTap service module on Cisco’s MDS 9000 series Fibre Channel switch to split writes at the SAN level for replication. This method of splitting writes has suffered at times from connectivity problems, which can cause some servers to lose access to the SAN, Ferrara said.
The bank already has matching Hitachi Data Systems USP-V high-end disk arrays at both ends of the disaster recovery network, but Ferrara said he’s hopeful that using Zerto to split writes for replication at the hypervisor layer will allow for more efficient use of storage — and eliminate the SANTap issues.
Other users just starting to test Zerto still have questions about the resiliency and efficiency of replication over IP-based networks between hosts, as opposed to the Fibre Channel-based networks used between arrays.
“I’m really used to Fibre Channel, even across sites,” said Ed Czerwin, a systems engineer for a large medical devices company near Zurich, Switzerland.
Czerwin’s company currently has Fibre Channel networks that use EMC’s Symmetrix Remote Data Facility replication to send a large amount of data — about 20 TB — over the wire daily. Fibre Channel is a lossless protocol, while IP networks see the occasional dropped packet. But there are third-party WAN accelerators that can help with these issues on IP networks, and Czerwin said he’s interested to see if Zerto will integrate with those tools.
“I consider IP to be a less resilient protocol than Fibre Channel, and maybe it’s my old school upbringing, being a SAN guy before I became a virtualization guy, but I just really want to see how fast [Zerto] can do [replication] and how efficient it is,” he said.
Meanwhile, the timing of Zerto’s emergence from stealth is interesting, given the impending release of vSphere 5, which will support host-based replication through Site Recovery Manager (SRM), according to officials at VMware’s Partner Exchange conference. With SRM for array-based replication, VMware has relied on partners to perform the actual data movement, rather than creating its own replication product. It remains to be seen whether the upcoming feature in vSphere 5 will work the same way, or if it will directly compete with what Zerto is doing.
VMware certification is also an outstanding question for Czerwin. “The main question I have is, do they have plans in the works to be certified as VMware supported, and do they have plans to get other third-party products certified for it? In my mind, that’s what makes or breaks a startup.”