LAS VEGAS -- VMware and Cisco are working on ways to improve virtual machine mobility, especially over distance. Moving virtual machines around and among dispersed data centers is a big issue for IT pros.
At VMworld 2011 on Tuesday, the two companies talked
IT managers at the show welcomed the news.
“It’s pretty slick,” said Bob Plankers, a virtualization architect at a Midwestern university. “Think of it as tunneling layer 2 over layer 3,” he said. With VXLAN, “your internal network is all of a sudden up in the cloud,” which could be especially useful for disaster recovery.
VXLAN extends the venerable VLAN technology, which Cisco introduced 18 years ago, said Soni Jiandani, Cisco senior vice president for server, access and virtualization technology in a VMworld session.
“It is clear that VLANs have taken us as far as we can go,” she said.
Indeed, the problem of VM mobility over distance is a hot topic at the show and among IT managers, who want to move workloads between data centers for load balancing and business continuity purposes, or to move applications into the public cloud. But while advances in virtualization, storage replication and automation have come a long way, the network and application layers remain an obstacle, one IT manager said.
“What we are missing is technology at the application level that allows the VM to run in active/active mode,” said an IT director for a large computer equipment provider.
Virtualization and storage technologies such as EMC VPLEX make it relatively easy to move VMs between data centers, but that is only part of the problem. Applications that have been moved often no longer work because their IP addresses are no longer correct and need to be reassigned. Presumably, VXLAN will fix that.
It’s unclear if and when the IETF will accept the VXLAN standard. In the meantime, IT managers are exploring other techniques for achieving VM mobility over distance.
One approach is to stretch VLANs across layer 2, but that’s a poor solution today, said Sanjay Aiyagari, a VMware partner architect speaking at a session on designing networks for multi-site vSphere deployments.
“VLAN stretching works in the metro, but as you extend it through multi-sites, IP addresses are no longer topologically correct,” he said.
Another approach is to proxy workloads behind a load balancer. “You want to apply [networking] policy at a [layer up] from the application,” Aiyagari said.
Relying on a load balancer can, in some instances, introduce latency, and requires the purchase of additional infrastructure, but it can prevent a lot of pain and suffering when it comes time to move VMs, said Simon Hamilton-Wilkes, a solutions engineer at F5, which offers its local and global traffic managers for this purpose.
“You want to keep IP addresses the same, or it can be a real nightmare,” he said.
Check out our full VMworld 2011 conference coverage.