Microsoft postpones live VM migration for Hyper-V two more years
6:30 pm EDT September 9, 2008 - Late this afternoon, VMware sensed an opening in Microsoft's virtualization strategy, and it has struck back with an early launch of its VMotion promotion -- touting the one feature its server-based virtualization products will continue to have over Microsoft until at least 2010.
Responding directly to Microsoft's statement to BetaNews earlier this afternoon, VMware spokesperson Melinda Marks told us, "Live migration has nothing to do with unplanned downtime. For unplanned downtime, VMware has VMware HA. Live migration solves a completely different set of business issues and is where Microsoft offers a solution called Quick Migration. VMware VMotion (live migration) allows businesses to perform planned maintenance and dynamic load balancing across an entire cluster without any end-user disruption.
"That's why live migration is so powerful -- it is transparent to the end-user," Marks continued. "It as an enabling technology that allows IT administrators to perform routine maintenance during normal business hours instead of scheduling it for evenings or weekends. It also allows administrators to treat individual physical servers as a logical pool of resources. So now companies don't just increase utilization on individual servers, they increase utilization across the entire cluster of servers. Microsoft Quick Migration causes end-user disruptions during a VM migration so IT admins are back to scheduling maintenance windows at 10 pm or Saturday morning. Microsoft knows this is a serious deficiency of its offerings and that's why it tried to show they will have it in the future. (We came out with VMotion in 2003.)"
2:15 pm EDT September 9, 2008 - In a dramatic reversal of Microsoft's apparent stance not even 24 hours ago, when it demonstrated its future live virtual machine migration feature for the first time while announcing a delay in its implementation until 2010, a Microsoft spokesperson told BetaNews this afternoon that the live migration feature may be required by only a small subset of customers. The spokesperson then added that it may also be undesirable for data centers, and even a potential obstacle to a "high availability strategy."
"Live migration is a useful technique for load balancing servers, and when used in conjunction with a high availability service to keep applications and services continuously running when something goes wrong," the spokesperson told BetaNews. "We've found that the early adopter customers of live migration use it on a very limited set of servers."
The spokesperson went on to reference German printing press manufacturer Heidelberger Druckmaschinen AG, which had been a VMware customer until it became an early adopter of some competitive Microsoft products, including Hyper-V. As Microsoft's spokesperson told us, "Heidelberg uses VMware's live migration feature for high availability on about 10 servers, and they have more than 400 physical servers. For them and most other businesses using server virtualization to consolidate servers, live migration isn't a required technology.
"If you're virtualizing today without high availability then you should re-evaluate that strategy," the spokesperson continued. "Virtualization and high availability go hand-in-hand. Windows Server 2008 Enterprise and Datacenter editions provide Hyper-V and integrated Failover Clustering at no additional charge. In the case of unplanned downtime, VMware ESX Server can't live migrate because there is no warning. Instead you must have VMware HA configured and the best it can do is restart the affected virtual machines on other nodes, which is the same as what is provided with Windows Server 2008 Hyper-V and integrated Failover Clustering."
9:08 pm EDT September 8, 2008 - At a virtualization product launch today, Microsoft give a long-delayed demo of Hyper-V live migration, but then went on to slate the feature's eventual release for the next edition of Windows Server.
In showing the upcoming capability to a crowd of customers in Bellevue, WA, Bob Muglia, senior VP of Microsoft's server and tools business, suggested that in the Windows Server product which follows Windows Server 2008, users will be able to instantly migrate virtualized software deployments from one server to the next, for consolidation on the fly.
But the live migration demo may actually come as bad news to some data center admins, who have been looking forward to Microsoft adopting some form of live migration since 2006. Microsoft's first delay of this feature was announced 16 months ago, after the company had promised it for "Longhorn," which became Windows Server 2008. The feature was cut, said product managers at the time, in order that Hyper-V could meet its launch window; but then that window was later scooted to 90 days after Windows Server 2008's own launch.
A little over one year later, the feature may now be waiting for a launch date two years from now, at the earliest. Without mentioning a specific time today, Microsoft stated live migration would be ready for Windows Server 2008 R2, which it touted as the very next version of the server operating system. Yet this product roadmap updated last month clearly marks the R2 version as "scheduled for release during 2010."
Live migration may actually be a necessity for some data centers, especially for systems that use failover clustering. If a virtual machine is running on a system that fails, conceivably a former state of that machine could be restored from a backup, but that would take time. With live migration, data centers can relocate running instances of critical servers between physical processors, with zero downtime.
In the absence of live migration, this year, Microsoft had been trying to promote an alternative scheme it calls quick migration, not only as a substitute but possibly -- if you follow the logic -- something even better. "Using Windows Server Hyper-V and the quick migration capability," reads a January 2008 Microsoft white paper, "you can consolidate physical servers and, at the same time, maintain the availability and flexibility of business-critical services during scheduled maintenance, or quickly restore services after unplanned downtime."
With the quick migration scheme, Hyper-V can rapidly save the state of an active machine to disk storage, reassign that saved state to a new physical processor or cluster, and then re-activate that state without loss of activity, in a process that Microsoft describes as "minimizing downtime."
Minimizing, though not eliminating. The company's own engineers admit that, assuming everything else goes perfectly, downtime could be cut to as short as six seconds.
Though quick migration was billed as a new feature by Microsoft at the time the term was introduced, last year, a technical representative of Microsoft Ireland admitted straight out that quick migration was actually a renamed feature that appeared in Windows Server 2003, known then as host-level clustering.
So rebuttals of Microsoft's arguments by VMware and its supporters have typically referred to the feature by its original name, especially in comparison to VMware's VMotion live migration.
"Without live migration, any migration solution still involves downtime, and therefore can't really compete with VMotion. Host-level clustering is great, and something I'd love to see VMware tackle for those shops where VMware HA just isn't quite enough to meet SLAs [service-level agreements]," wrote IT administrator Scott Lowe in July 2007. "But host-level clustering is no substitution for VMotion."
In a series of April 2008 blog posts for Microsoft's virtualization team, lead program manager Jeff Woolsey delicately implied that Hyper-V's quick virtualization was actually more fail-safe than VMware's live migration, by giving administrators a picture of a single point of failure, and implying that quick migration avoids that fate.
"Virtualization actually creates a major problem: single point of failure," Woolsey wrote in Part 1. "Think about it. In the past, you may have been running 20 workloads each on their own physical server. When one of those servers goes down it's bad, but probably not the end of the world. In a virtualized environment, suppose you have 20 workloads running a top a single server. What happens when that physical server goes down? All 20 workloads go down. That's not bad, that's catastrophic."
In Part 2, Woolsey followed up with this: "In the case of unplanned downtime, VMotion can't live migrate because there is no warning. Instead you must have VMware HA configured and the best it can do is restart the affected virtual machines on other nodes which is the same as what is provided with Windows Server 2008 Hyper-V and Failover Clustering."
VMware, meanwhile, already has a live migration feature, which has been one of its major selling points to data centers -- and apparently will be for at least two years to come.
Next: Microsoft finally solves the Virtual Machine Manager debacle...