Network puts Windows Server 2008 to the test

We tested Windows Server 2008 Enterprise Edition RTM on a switched Gigabit Ethernet D-Link network using primarily a Dell 1950 server equipped with a dual quad-core 1.6GHz CPU, 32GB of dynamic RAM, two Broadcom Gigabit Ethernet network interface cards connecting it to the network and a QLogic 2GB Fibre Channel card connecting it to our internal storage-area network.

To assess operating system compatibility, we also ran Windows Server 2008 on other servers including an HP585G2 with four, dual-core AMD Athlon 64 CPUs with 4GB of DRAM per card and Compaq NetIntelligent Gigabit Ethernet NICs; three Compaq DL-140 boxes (each with dual Intel Xeon CPUs running at 3GHz). The client machines used in testing included various HP and Apple workstations and notebooks running Windows XP SP2, Windows 2000 Professional, Apple 10.4.11 or 10.5.1.

We imported 100,000 users into the Active Directory using LDIFF, creating 32 groups, and seven administrative domains. After importation, we randomly checked each of the Active Directory components for signs of an accurate importation; it was successful. We then made various changes to users, groups, and policy objects and watched for the changes to appear in the new Windows 2008 AD audit logs. Despite the comparative size, changes were fast.

In terms of performance testing, we ran a series of tests where files and folder were copied using four HP PCs, each formatted with fresh copies of Windows XP SP2 first and then with fresh copies of Vista SP1. We copied identical folders containing 60MB of files (a mixture of files ranging from 1K to 17.2MB) on our Dell 1950 PowerEdge server (no antivirus software was used on clients or server) using a batch file, noting copy execution times. Vista SP1 copies were significantly faster over the Ethernet 10Base-T hub that we connected the PCs to the server with (emulating slower speed links). We also compared streaming times using MP3 files from each server.

We also tested performance between NDIS2 (Vista) and NDIS1 (Windows XP, and various SAMBA clients) for file/folder copy speed performance, and found NDIS2/VISTA clients are strongly favored in busy networks.

We measured simple disk access through file/folder copying on a server (rather than client-to-server) to be the same between Windows 2003 and Windows 2008 Server Editions. We tested iSCSI host and target software, and found it to have the same speed as R2-supported file I/O.

We also tested IIS 7 via Web get/posts using inclining concurrency of the get/posts (using static pages), and found that there was no additional performance advantage of Windows 2003 over 2008 server editions until concurrency (number of users getting and posting) was raised significantly, where Windows 2008 and IIS 7 became very fast compared with Windows 2003 and IIS 6. Both versions of IIS were running in default/unoptimized configurations.

Logon compatibility was checked between Windows 2008 Enterprise Edition with Windows 2000 Professional, XP (SP1 and SP2), Vista (with and without SP1,which was released on the same day with the RTM of Windows 2008 Server editions), MacOS 10.4.11, 10.5.1, Ubuntu Linux, Red Hat Linux Enterprise 5, and Debian Sarge. We also used Windows 2008 Enterprise Server (six instances) as a VM guest operating system under VMware ESX 3.5 with no issues.

We discovered that support for Macintosh/Linux in advanced services such as Network Access Protection is absent (and can be handled as an unprotected exception), and other Macintosh services (formerly known as Microsoft Services for Macintosh) are no longer supported; these included AppleTalk cross-platform support, and secure logon (encrypted with hash using differing access methods).