A telecommunications firm in Britain is supporting its Internet subscribers using the Ubuntu Kernel-based Virtual Machine (KVM) hypervisor, open source management software from Convirture and an approach to virtual machine clustering that shuns live migration.
Between 26 and 40 Ubuntu 10.04 LTS virtual machines run on two geographically separate clusters to support the ISP division, depending on dev and test activities, according to Linux administrator Jamie McDonald.
Ubuntu and KVM support applications and services in JT Global’s ISP division, including RADIUS authentication, the Puppet configuration management tool, TACACS authentication for routers, NFSen networking traffic analysis, NetFlow data analysis, BackupPC for host-based backups, and some homegrown Web-based applications for IP address management.
Ubuntu 10.04 LTS is the default OS in the ISP division’s environment, and KVM seemed the logical choice when the company began to virtualize five months ago. Ubuntu, along with most of the open source world, has moved away from Xen in favor of KVM as the default open source hypervisor, and VMware’s vSphere was deemed too costly.
While most of JT Global’s ISP division applications and infrastructure are homegrown, it chose a pre-built open source server virtualization management offering from Convirture to provision and manage virtual machines instead of using Ubuntu’s native command-line utility, libvirt.
“From a provisioning point of view, not everybody else here is familiar with working on a command line,” said McDonald. “If they need a virtual machine, they can create one nice and easy without needing to know how the back end fits together.”
The company chose ConVirt Open Source after investigating open source tools such as OpenQRM and vSphere.
Once again, cost was an issue when considering VMware, and ConVirt Open Source’s interface was the best fit for the types of services JT Global was looking to support, McDonald said. “It pretty much came down to OpenQRM and Convirture. OpenQRM seemed…aimed more at the commercial hosting market and the provisioning of virtual private servers.”
Clustering approach shuns live migration
The ISP’s dual-host clusters are located about five miles apart, each attached to their own storage systems. JT Global prefers to failover between clusters for added redundancy, rather than pooling all the resources into one cluster with live migration.
“The two clusters are unaware of each other – there’s no centralized shared storage between the lot, and we did that for redundancy,” said McDonald. “We’ve got two completely independent [clusters], and that means we can set up clustering on a per-virtual-machine basis, rather than having one system where everything is automatically failed over. It gives us more control over taking services up and down.”
The firm has skipped ConVirt’s storage provisioning features, preferring to deliver its own storage on another Ubuntu server in each of its clusters. The storage is attached via iSCSI using software from the OpeniSCSI project.
Every night, shell scripts developed in-house take snapshot backups of the virtual machines and are replicated using rsync to a NAS device at the other site. JT Global keeps copies of all its virtual machines at both sites, along with duplicate networks with identical bridge configurations. If a machine needs to be moved, it’s failed-over from one node to the other building.
McDonald said he would prefer not to script snapshots, but full-VM backups are only supported in the costlier Enterprise edition of Convirture’s offering. Convirture Open Source only snapshots memory of the virtual machines. “Some kind of built-in backup service would be cool, instead of us having to write our own scripts and take backups every night.”
Beth Pariseau is a senior news writer for SearchServerVirtualization.com. Write to her at email@example.com.