For the past two years, Unisys has been tracking uptime on installed ES7000s, and says that the machines are as reliable as big Unix boxes, and in some cases even better.
Reliability and availability are key factors in any IT server decision when mission-critical applications are concerned. Business is lost when machines go down and take applications with them. Windows NT was not exactly known for stability, compared with Unix, so with Windows 2000 Microsoft did a lot of work to improve the availability of the boxes. Server makers like Unisys also compensated with their own autonomic and recovery technologies to fill in gaps. With Windows 2003, Microsoft says, it did a much better job making the blue screen of death more of a memory than a daily occurrence. But the proof is in the pudding, or, in this case, in the statistics. It is one thing to talk about reliability, but it is quite another to demonstrate it.
That is what Unisys set out to do by watching how ES7000 servers running Windows Datacenter Edition actually performed as customers were implementing and running their applications. Mark Feverston, vice president of platform marketing for enterprise servers at Unisys, says that, two years ago, the company chose a sample of 68 servers from its installed base to watch. These were all production systems, not test or development machines, and they all had the Sentinel Health Advisor monitor activated. From the moment the implementation of applications began, Unisys kept track of uptime. They ran a mix of Oracle, SQL Server, and DB2 databases.
Unisys found that, in a stand-alone configuration, over the course of two years, the ES7000s running Datacenter Edition provided an average of 99.9% system availability in a stand-alone configuration. That’s with no high-availability clustering. The conventional wisdom from a few years ago is that a mirrored Windows NT 4.0 server (with high availability clustering) provides about 99% availability and a stand-alone Windows NT 4.0 server offers about 97% uptime. In the pool of Unisys customers, the machines running Windows 2000 Datacenter Edition had about 20 to 40 minutes of downtime a year, while installations running Windows 2003 are averaging what will work out to just under 22 minutes of downtime a year. Windows 2003 is hitting about 99.996% uptime on the ES7000s. Unix servers and mainframes can get to the Holy Grail of five-nines availability (about 11 minutes of downtime a year), but they usually have to be clustered to do it. The conventional wisdom is that a big Unix box can do three- or maybe four-nines of availability in a stand-alone environment.
Feverston says that most of the downtime customers experienced on the ES7000s was due to their using unqualified drivers on the Windows platforms, or had to do with the implementation process in the first few months of the installation of the Windows-ES7000 setup. This downtime did not, by the way, include planned downtime for maintenance or upgrades, nor did it include unintended shutdowns that were not related to the operating system or environmental problems. The uptime is Windows uptime, not application uptime. Of the 68 systems in the survey, Unisys said, 53 machines had 100% operating system uptime (again, this is not including downtime from maintenance and upgrades, etc.), and that only four machines in the pool reported less than four-nines of availability in a year.
Microsoft co-sponsored the reliability study initially, but when Windows 2003 was launched, last April, Unisys had to continue gathering data on its own. Feverston said that the original intent of the study was to help Unisys engineers identify ways of improving uptime, but having seen the numbers, the company is going to use them to further promote the ES7000 running Windows in the data center. There is no reason anymore for customers to go with proprietary RISC/Unix platforms, says Feverston. These numbers are right up there with the best of any machine in the data center.
