I’ve discovered that Hyper-V virtual machines run across a purposefully-pitiful iSCSI connection can run faster than those which are hosted on local disks. To explain, let me first outline my components:
- Two Toshiba Satellite M305D-S4830 laptops, each with 4G of RAM and dual x64 processors. Nice laptops, actually, for the $570 each I paid for them. I hear they’ve got 40 more of them for sale at the Denver MicroCenter, but I digress…
- For disks, the laptops are connected to another server in the basement. That server runs Rocket Division’s Starwind iSCSI Target software. That server currently hosts two LUNs, one 512M LUN for the quorum with a second 45G LUN for a single virtual machine.
- The laptops are connected to their iSCSI data store through 100Mb/sec connections. There are three hops between each laptop and its iSCSI data store, each through a consumer-grade low-end hub or switch.
- Of those three network devices, one is a Vonage router turned around backwards to be used as a hub. The second is a Linksys WRT54G of which its built-in four-port hub is being used. The third is a D-Link DGS-2208 gigabit switch. Each laptop hops through each of these three devices in order to get to their data store.
- Each laptop has Windows Server 2008 installed with the Failover Clustering feature and Hyper-V role installed. There is only a single NIC in each laptop. That NIC, bound to multiple IP addresses, is simultaneously hosting the role of production network and iSCSI connection.