GeoCluster - Windows from here to Timbuktu
Double-Take Software - which was founded to provide data replication for servers and their applications - has been transforming itself into a workload optimization and high availability software vendor. And like everyone else, Double-Take is trying to get into the virtualization game - not just to make money, but because servers are being virtualized and their virtual machines need a lot of the same resiliency that was originally provided for physical servers.
Like other geographic replication tools, Double-Take's GeoCluster software uses an IP network link to link source and target servers together, replicating data asynchronously. With this asynchronous link, servers can be geographically dispersed, and data replication can still happen over relatively-low-bandwidth, long-distance links - usually T1 lines among Double-Take's customer base.
GeoCluster, which runs atop Windows Clustering Service, adds a few tricks of its own. Cluster Service can span up to 16 nodes in a cluster, but requires the nodes to be linked to a storage area network. But using the data replication software that the company sells as a standalone product for Windows and Linux servers, GeoCluster allows each server to keep its own local copy of the data, which means you don't need to have identical servers in the cluster and you don't need to buy a SAN. Windows Cluster Service also is not made to span long distances. That is a function that Double-Take (as well as a number of other vendors) slap on top of Cluster Service to make some money.
Even with that, Cluster Service made it tough to do meaningful separation for servers, which you want to truly protect your data and applications. According to Bob Roudebush, director of solutions engineering at Double-Take, Windows Server 2008 has two important tweaks that make geographically distributed clusters more practical compared to setups running on Windows Server 2003, which had some issues. And the new GeoCluster Version 5.1 release from Double-Take takes advantage of them.
The first limitation in the clustering services that come with Windows Server 2003 was that nodes separated from each other geographically in a cluster still had to be in the same subnet on the corporate network. This forced network administrators to stretch a subnet address over the wide area network. This was both tricky and annoying. The heartbeat inside cluster services for Windows Server 2003 was hard-coded at 500 milliseconds.
This was problematic because depending on network traffic and distances, the latency between source and target machines could be in excess of 500 milliseconds, and when that happens, the replication link is broken. This limits geographic separation in most customers to on the order of miles and kilometers, which defeats the purpose of setting up a GeoCluster in the first place.
With the cluster services inside Windows Server 2008, Microsoft now lets the two machines linked together be on different subnets, and it also lets administrators set the timeout on the heartbeat. These changes make GeoCluster for Windows customers more relevant. Zane Adam, senior director of virtualization strategy at Microsoft, says that roughly one-third of the companies deploying Windows Server 2008 thus far have deployed it in clusters for high availability.
GeoCluster 5.1 is also aware of virtual machines running atop the Hyper-V hypervisor, and virtual machines can be replicated and failed over from one box to another, just like physical boxes do. GeoCluster is also aware of the quick migration feature that Microsoft puts in the Enterprise and Datacenter Editions of Windows Server 2008, which allows for a VM to be teleported around a network of machines that have shared storage. As with physical machines, the combination of GeoCluster, Hyper-V, and quick migration means that a SAN is not necessary. All the servers can use local storage because of the replication that GeoCluster is doing behind the scenes.
GeoCluster 5.1 supports Windows Server 2003 and Windows Server 2008, although given the limitations of the earlier Windows, this is obviously not where either Double-Take or Microsoft is focused. (Windows Server 2000 is not supported). Each node in a cluster needs to have its own GeoCluster license, which costs $4,940 a pop.
While Double-Take has data replication and high availability failover products for Linux, Roudebush says that at this time the company no plans to offer GeoCluster on any Linux. The commercial Linux distributions from Red Hat and Novell have had support for clustering for years, but the Double-Take for Linux product, which provides replication and failover on closely packed servers, only supports Red Hat Enterprise Linux 4 and 5. ®