Similar servers were used to test each of the data protection packages. To support the backup server and media server roles that are part of both the Symantec NetBackup and CommVault Simpana data protection configurations, two Cisco UCS C200 Servers were used with 10Gbit converged network connections. In the test configuration both the NetBackup and Simpana servers handling media also had a data deduplication option installed. The Avamar configuration, however, required only an Avamar Data Store Single Node with a 1Gbit LAN connection to the converged network. In addition, the Avamar and Simpana configurations included a proxy VM using VMware’s SCSI hot-add to back up data, which was transferred over a LAN connection to the media server or data store.
The Avamar VM proxy client is packaged as a preconfigured OVF appliance with two virtual cores and 2GB RAM. The CommVault Virtual Server iDataAgent needs to be installed on a VM running a Windows server or desktop OS.
The Symantec NetBackup media server, on the other hand, implements backup directly over a SAN fabric. All backup tests for NetBackup utilized a LAN-free configuration and were run exclusively with direct SAN access to the VI disks. Nonetheless, significant network traffic was measured between the media and backup server during a VM backup.
Real World Business Continuity
A corporate email system is deeply embedded into IT infrastructure and the effectiveness of the EMAIL application portfolio dramatically impacts the productivity of the entire corporate staff. That’s why the stability, reliability, and security of a corporate email system has far-reaching implications for IT and LoB organizations alike. As a result, multiple detailed backup and restore tests were run on the VMs hosting Exchange 2010 to investigate the efficiency and performance of backup and restore operations needed to be performed in a business continuity plan for an email system.
Each member of the Exchange DAG configuration was provisioned with a 40GB system disk and a 150 GB maibox database volume. One member was designated as the active member for production. The second member maintained a passive copy of the mailbox database and was used in all of the backup and restore tests. The Microsoft Load Generator Tool (LoadGen) was used to generate consistent measurable amounts of email data in the mailbox database on the active server. With LoadGen, specific backup and restore test scenarios were readily created.
Backup testing began with a baseline full which initializes CBT for a VM. Running a full backup without CBT is a distinct one- time event for each VM with Avamar,as the 5.8XAvamar disk-block store always makes a CBT-based backup a full backup. With a traditional backup architecture with distinct full and incremental backup files, periodic backups with out CBT or synthetic full backup processes will be required.
In any backup with a large volume of data, such as an initial full backup, a SAN-based VM backup has a distinct advantage over an appliance, which relies on VMware’s SCSI hot-add to access the logical disks belonging to the VM being backed up. In a SCSI hot-add scenario, backup data must traverse the appliance’s SCSI and network protocol stacks, in addition to the hypervisor’s SAN stack to get to the backup server. For a backup solution supporting both SAN- and appliance-based backups, a SAN-based backup of a VM with a large volume of data is typically twice as fast.
Nonetheless, even with an inherent advantage in topology for the initial backup of the VM hosting Exchange, Symantec NetBackup exhibited a remarkably small 11% advantage in backup time over Avamar. Moreover, In a CBT-based backup of the VM after a modest 2% data change in the mailbox database, Avamar was 4.8 times faster than NetBackup and 3 times faster than Simpana.