In the music industry, people talk about that "difficult second album," but when you're introducing virtualization technology into your data center it's the last 20% of workloads that tends to be the most difficult. This tends to be those mission-critical applications that are hardest to virtualize, and they're the ones you need to be the most cautious about when making changes to the way they are run.
But those workloads are being virtualized more frequently these days, and one company that's benefiting from this trend is Stratus Technologies. The company is as old as the hills (well the 1970s anyway), and it specializes in fault-tolerant servers. Essentially, these servers are two machines built from industry-standard parts, with Stratus' own ASICs (application-specific integrated circuit) to provide lockstepping.
They offer 99.999% uptime for VMware 5.1, the company claims, and to prove the point (and as a marketing gimmick), Stratus offers a $50,000 zero-downtime guarantee for newly purchased systems that covers both server and hypervisor for six months from the time of installation.
Historically, about 10% of the company's business has been selling machines to act as VMware hosts. But in the last year or so this has shot up to as high as 80% for some of its offerings, as an increasing number of organizations virtualize their Oracle databases and the like. Denny Lane, a Stratus director, believes the reason for the surge in interest for fault-tolerant servers for VMware hosts stems from the fact that VMware's High Availability software is far from perfect.
That's because in the event of a hardware failure, you still experience downtime if a system crashes and restarts on a new system, and any data "in flight" will be lost. "VMware has convinced people that HA means no down time, but that's a myth," says Lane.
He also argues that a single fault-tolerant Stratus server with internal storage costs less than buying the two or three standard servers you'll need to use with VMware HA, along with shared storage and networking between them all. (As a guide, a basic fault-tolerant server starts at around $15,000, and Lane reckons that a single Stratus server with internal SAS disk storage can work out 30% cheaper than running three HP servers and a SAN, or 10% cheaper than similar configuration with Dells.)
Of course, a fault-tolerant server protects you against hardware faults, but it doesn't help you when the OS inside a VM crashes. That's why in larger data centers Lane advocates the belt-and-braces approach of running fault-tolerant servers and VMware HA for those difficult mission-critical applications that your business may depend on.
That way you don't have to worry about hardware faults bringing down your virtual machines, while VMware HA monitors your VMs and restarts them in the event of an OS crash.
But it's in smaller businesses, remote locations and businesses new to virtualization technologies with few if any VMware skills that fault-tolerant hardware really comes to the fore, according to Lane. That's because VMware HA goes against the whole philosophy of virtualization technology, he believes.
"To do VMware HA you need to have three or so servers," says Lane. "But if you can virtualize all your workloads onto a single server, then having three servers for HA undoes all the benefits of consolidating onto a single server."
Of course, Lane istrying to sell fault-tolerant computers after all, but you can't help thinking that his last point is a good one.
Paul Rubens is a technology journalist and contributor to ServerWatch, EnterpriseNetworkingPlanet and EnterpriseMobileToday. He has also covered technology for international newspapers and magazines including The Economist and The Financial Times since 1991.