Back up without crashing

As more servers are virtualized, backing up and protecting them becomes more of a problem. It's not enough for IT to back up each virtual server and its data. Protection is needed for the virtual server's image -- its OS, configuration and settings -- and the metadata on the physical server that identifies the virtual server's relationship to networked storage. The challenge is to choose the right virtual server backup option:

-- Traditional agent-based backup software, which installs a software agent on each virtual machine to back it up.

-- Serverless backup, which offloads backup processing from virtual machines (VMs) to a separate physical server.

-- Snapshotting VMs with software included with the virtualization package to protect data and images.

-- Writing scripts and executing them to quiesce (minimize the number of processes running on) the VM, back up its contents and restore the VM.

-- A combination of agent-based software and cloning.

Each virtual-machine backup approach has its advantages and disadvantages. Chief among the disadvantages is the effect on network performance and utilization. While virtualization can result in better utilization of server resources, backing up all the newly created VMs concurrently for a physical server can overwhelm the network and take resources from applications running in other VMs.

By virtualizing physical machines you increase the number of servers contending for one bus. So Chris Wolf, senior analyst, Burton Group, suggests users only virtualize servers that contain a PCI-Express (PCI-e) bus.

"When you have a shared I/O channel for all your PCI devices, traditional PCI devices can severely slow you down when you talk about six to 10 VMs sharing the same bus," Wolf says. "PCI-Express should be the bus of choice for all new virtualization deployments, as it offers a transfer rate up to 16Gbps in full duplex, compared to PCI Extended, which has a maximum throughput of 4Gbs."

Also to consider is the cost associated with agent-based backup software used in a virtual environment. Since most vendors of backup software require a separate license for each backed up VM, as well as one for the physical machine hosting the VMs, licensing costs can increase quickly.

The advantage of agent-based backup software is that IT administrators are familiar with it, having deployed it for many years to back up the physical machines in their environment.

What Users Say

Increasingly, IT users are opting for a combination of methods to back up virtual servers. A common approach is to use agent-based and serverless backup for protecting data on VMs, combined with cloning or snapshot technology for protecting and recovering server images if hardware fails.

One user who has adopted this combined approach is Jim Klein, director of information services and technology for the Saugus Union School District. Klein uses the Open Source Xen virtualization hypervisor to virtualize the blade servers in his environment.

"We treat VMs like any other server by using backup agents from Bacula, an Open Source backup solution," says Klein.

While Klein uses agent-based backup to protect the data on his VMs, he uses cloning technology to deal with server failures.

"For the base virtual machine images, we store them on the host computers and replicate them to the failed server or store them on a network attached storage, or NAS device." The metadata for Xen, which describes how servers attach to storage resources, is stored in a database called XenStore, which is included with the Xen hypervisor, and can be backed up easily by simply copying files to a backup device.

Art Beane, IT enterprise architect at IFCO in Houston, also has found that a combination of backup technologies works best for him. Beane uses NetApp's SnapManager software to snapshot the data on his NetApp SAN, and cloning to back up the servers attached to them.

He has six physical machines virtualized with VMware Infrastructure 3 into 23 VMs.

"Our backup plan is common to virtual and physical servers," Beane says. "No persistent data is allowed on a server, only on the SAN. The SAN gets a snapshot backup every two hours and a full backup daily."

The system drives -- both physical and virtual -- in Beane's servers get imaged weekly using Acronis' TrueImage. In the event of a catastrophic server loss, the Acronis image can be restored either to a physical box or as a VM.

This multilayered approach is the configuration Wolf most often recommends.

"For smaller dedicated application servers, running the agent inside the VM is certainly ideal," Wolf says. "That needs to be combined with a policy and change-control process for the creation of storage and snapshots as well."

In this way, when it comes time to recover files, all an IT admininstrator needs to do is bring online the backup copy of the VM snapshot and then restore the latest data files from the agent-based backup.

Consolidation Play

Another method for backing up VMs is to use serverless or consolidated backup technology. In consolidated backups, backup processing is offloaded from the VM and physical server to a separate backup server called a proxy, thus helping to avoid any performance impairments.

Consolidated backup is commonly deployed using a combination of VMware Consolidated Backup (VCB) and agent-based backup. VCB consists of a set of drivers and scripts that enable LAN-free backup of VMs.

In a consolidated backup, a job is created for each VM and that job is executed on the proxy server. The pre-backup script takes a VM snapshot and mounts the snapshot to the proxy server directly from the SAN. The pre-backup script also quiesces the Windows NT file system inside the VM. The backup client then backs up the contents of the VM as a virtual disk image. Finally, the post-backup script tears down the mount and takes the virtual disk out of snapshot mode.

Taking snapshots and cloning VM images has many advantages. Like agent-based backups, most IT administrators are familiar with them. Snapshot and cloning capability also is included in many virtualization packages such as VMware and XenSource, and with many traditional backup tools.

"There's not a one-size-fits-all solution," Wolf says. "When you are dealing with large amounts of data such as in databases ... I prefer to configure VMs to use a raw LUN (logical unit) so the VM is not using a virtual hard disk, but actually mapping to actual storage resources on the SAN. That opens up the flexibility to use serverless backups, some of your snapshotting agents and all of the capabilities of your backup software that exist in the physical world."

This story, "Back up without crashing" was originally published by CIO.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2008 IDG Communications, Inc.

SD-WAN buyers guide: Key questions to ask vendors (and yourself)