The idea behind the Stratus 6200 fault-tolerant ftServer is to provide a device that performs as well as a similar server thats not fault-tolerant. This is no small task, if only because the process of providing continuous availability can soak up a lot of CPU cycles, which in turn slows things down. Stratus, with its dual quad-core Xeon processors, has solved this problem.
The Stratus 6200 ftServer was developed jointly with NEC. It sells for just under $50,000 in the configuration I tested, although less expensive versions are available. The server is available through a variety of resellers, including Dell, and began shipping in fall 2007.
Whats most interesting about the Stratus 6200 is the way it solves the overhead problem. The Stratus 6200 is, in reality, a pair of 2U servers that operates in what the company describes as “lockstep.” Both sets of Xeon processors perform the same instructions at the same time, the memory state in both machines is identical, and, if one system fails, the other picks up operations as if nothing has happened.
The pair of 2U chassis communicates through a backplane in much the same way that servers connect through a blade chassis. The servers can share power and network connections.
According to Stratus, the 6200 is aimed at organizations with the need for a lot of critical transaction processing. This includes companies in the financial services and manufacturing industries. At least one major ISP is also using Stratus servers, according to the company.
The 6200 comes with either Red Hat Enterprise Linux or Microsoft Windows 2003 Server. In either case, Stratus provides what it calls “hardened drivers,” to make sure that the operating systems handle the fault-tolerant aspects of the server operation. The version I tested came with Red Hat Enterprise Linux pre-installed.
Getting the server up and running was no different from getting any other Red Hat Linux server running. Stratus doesnt provide any special configuration package, nor does it provide any specific management software, except for the utilities to control the operation of the server failover.
There is a Web-based GUI on the out-of-band management port that will let a remote manager check status, perform restarts and the like. That GUI does not provide any broader management capabilities, however.
As youd expect with a server that is delivered with unadorned Linux, the 6200 is intended for life in the data center.
While you can buy this product in a pedestal form that is intended for use in offices and other occupied spaces, it isnt necessarily a good idea. During tests, I found the server to be too noisy for use in spaces that have workers nearby. Using a sound-level meter, the 6200 measured 78 decibels in normal operation. According to a report by the Smithsonian Institution, this is equivalent to the noise on a busy street corner in Manhattan at rush hour. No matter how you compare it, this is one noisy server.
I reviewed the rack-mount version of the server, which was enclosed in a cage that also contained the backplane. To install this server into your data center, youd mount the cage (which is a lot lighter than the servers) and then slide the servers into place.
The backplane is designed so you can hot-swap the entire 2U chassis while the other chassis keeps running. Once the replacement server is in place, the two machines will synchronize their operations, and youll be protected again. Included in each chassis is a three-disk array, each providing failover for the other. You can choose between SAS (serial-attached SCSI) or SATA (Serial ATA) disks.
Stratus Server Keeps Critical
I tested this server using a variety of Windows and Linux clients. The tests involved connecting to the server using SSH, Samba and NFS. I performed large file copies (the same 7.25GB image file folder Ive used in previous server reviews), and I performed backups with the server as the target.
Everything worked as it should, and there was no indication at the client that your target was a fault-tolerant server with four Ethernet ports that could pick up the load as needed any time.
Effectively, of course, you really have one Gigabit Ethernet port, with everything else on standby either for the other server chassis if it needed to take over or on the active server if one of the Ethernet links was lost. Unplugging power cords and Ethernet cables had no effect on the servers operations, as long as something was connected.
Based on my test, its clear that operations such as credit card transactions, warehousing and emergency call dispatching can continue without interruption even with the loss of one source of power or part of the networking infrastructure.
You cannot, of course, have a fault-tolerant server in a vacuum. The ability to failover when a component dies is critical, but so is the ability to know what failed, and to know what to replace to fix it.
The Stratus ftServer provides this information in two ways. First, it sends out a constant stream of e-mails giving its status at any given moment. With its default settings, this can be a lot of e-mails, but you can configure the remote management application to send only certain alerts to certain people.
In addition, the ftServer sends messages back to the customer support staff at Stratus. I first noticed how this worked when I started getting calls from tech support when I did something like shut down and restart the server. Since the review process results in a lot of restarts, I got to know the remote support people pretty well. Eventually, they figured out that this was part of a testing process and stopped calling after making sure the server was really OK.
Tech support at Stratus seems to be extremely knowledgeable, but highly specialized. This means that if you need to ask questions about both the server hardware and the networking setup, youll end up talking to two people.
In addition, the staff doesnt seem to be focused on providing operating system support, so youll need to make sure your IT staff really knows RHEL or Windows. On the other hand, the tech support staff will be able to talk to you in detail about the health of the hardware and the proper way to set it up.
One thing thats not really addressed by Stratus, but that remains necessary for a real fault-tolerant computer system, is access to independent sources of power and independent network infrastructures. And an independent source of power means two supplies coming from outside the building. Your high-availability server wont do you much good if you lose all power in a single event.
Check out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.