How We Tested Backup Software
Our testing environment consisted of a 100-Mbps 15-node LAN with Novell NetWare 4.11 and Microsoft BackOffice 4 servers. Our NetWare server was a Micron Electronics Vetix 2LXI with twin Pentium Pro 200s, 128 MB of RAM and 12 GB of disk space connected to the network by an Intel 100B 10/100 switching Ethernet adapter, and an Adaptec 2940 UltraWide host adapter connected to a Hewlett-Packard Co. SureStore DAT drive. The BackOffice server was a Compaq Computer Corp. ProLiant 3000 with twin 333-MHz Pentium IIs, 256 MB of RAM, a 14-GB Compaq SmartArray Level 5 RAID, a Compaq Netelligent 100BASE-T network adapter and Compaq Wide SCSI connected to our tape array. Additional network storage was provided by a Raidtec Corp. FibreArray JBOD with 72 GB. The primary tape array was an ADIC Scalar 448 tape library that consisted of four DLT 7000 drives, 48 slots, 10 mailbox slots and a robotic arm.

Each backup package was installed on the appropriate server, and agent and client software subsequently were installed on the workstations, which included a 400-MHz Pentium II IBM Corp. IntelliStation with 256 MB of RAM running Windows98 that we used to set up all administrative clients. We then attempted to install the ADIC Scalar tape array, the SureStore DLT, a Sony Electronics AIT drive and an Exabyte Corp. Mammoth drive. In cases where the software only supported server-based backup, we used only the ADIC and SureStore devices.

Once we had everything set up, we configured the tape software to perform a full network backup, several timed unattended backups, and a critical performance backup that attempted to back up a variety of open NT server files, from Exchange and SQL files to a variety of documents, programs and Windows system files.

During testing we looked at the broad criteria represented in the report card, focusing on the overall success of the backup package to handle our network and the critical tasks we set up. We spent time monitoring the backups from our administrative workstation, and attempted to change job and queue priorities on the fly.

Ultimately, we subjected each package to several scenarios that represented a slightly more demanding set of tasks than you'd find in the real world. We were glad to note that with the few exceptions noted in the main story, each package passed our tests.