Beware of unintended consequences. Take virtual machines, for example. They’re ideal for taming server sprawl, reducing power consumption in the datacentre and keeping IT management costs down.
But demand for server I/O skyrockets as multiple guest operating systems and their applications stress what one physical server can deliver.
Gary Orenstein, vice president of marketing at Gear6, says that by deploying a centralised network cache, such as his company’s Cachefx technology, IT managers can create VMs to their hearts’ content without worrying about performance bottlenecks.
Gear6 has two cache appliances: the G200, which can store 250GB of data, and the G400, which can pack 500GB. Both handle more than 250,000 I/O operations per second. In the works for next year are smaller and larger appliances, Orenstein says, with the biggest targeted at 1TB of cache. He says IT needs “a change of mentality” in thinking about network cache devices. Instead of looking at cost per gigabyte, he says, IT should start looking at cost per I/O.
Cost of Cachefx? It starts at $350,000.
Use Ethernet mesh for your I/O woes
Derek Granath, vice president of marketing at Woven Systems, agrees with the Gear6 folks that virtualisation is increasing I/O demand. But he argues that other datacentre problems, such as the growth of unstructured data in SANs, can also degrade performance.
His fix for multiple performance bottlenecks in the datacentre is to use industry-standard 10Gbit/sec. Ethernet technology, specifically Woven’s multipath mesh switches, to create thousands of potential paths for the data to flow along.
Granath claims that the switches are smart enough to know which of the many paths is fastest at any given moment. Packets transmitted by Woven switches are followed by a tiny probe packet that measures latency so the switches know where the slowdowns are. At the end of this month, Woven will release its TRX 100 switch with 48 server links and four 10Gbit/sec.
Ethernet ports. Pricing starts at $7,950.
Monitor the deluge of virtual servers
Another vice president of marketing, Ken Klapproth of Entuity, sees the VM phenomenon as indicative of bigger problems confronting IT: the number of “ever-expanding applications now using the network” and end users’ “thirst to connect to more data resources.”
The cure, Klapproth says, is network management software that goes beyond the features of point tools while avoiding the cost and complexity that he asserts come with larger-scale management frameworks. Not surprisingly, Klapproth suggests that Entuity’s Eye of the Storm does the job.
The 2008 release, which will be ready at the end of the month, offers advanced reporting modules, including at-a-glance graphical views of network conditions. Also, the software now has a configuration monitor that lets you save settings on Cisco network devices. It can alert you in real time when the configuration data has changed, and it then lets admins easily switch back to the previous known good state. The upgrade also includes a progressive discovery feature that lets you continuously snag data on any SNMP-capable device on your network — even those you choose not to monitor. Oh, and of course, the new release can now monitor the health of virtual servers, so you know the condition of each virtual machine.
Pricing for Eye of the Storm 2008 starts at $50,000.