When we received the latest releases of Caldera's OpenLinux and Red Hat, we thought it was time to check their performances compared to Windows NT's - especially in light of the recently disputed NT vs Linux benchmark performed by the Mindcraft performance-testing lab.
Using the latest revision of the IDG test centre's file and print benchmark, we looked at file and print performance on Red Hat 6.0 and Windows NT 4.0 with Service Pack 4. We had intended to look at Caldera's Open Linux 2.2 as well, but unfortunately Caldera could not recognise our AMI RAID controller.
Unlike artificial workloads, our benchmark uses real applications running realistic user activities to measure the overall performance of the network system. We tested on a Gateway quad-Xeon box with 512MB of memory and two network segments supporting a total of 30 Pentium II 266MHz clients running Windows 98.
The clients were each connected via a switched 10Mbps Ethernet, with each switch connecting to a 100Mbps uplink to the server. For file and print compatibility, we ran Samba on Red Hat. Samba is the file- and print-sharing service that lets Linux emulate or serve Windows NT and 9x servers.
For each test, we used a separate domain controller for log-on duties. We made the servers we wanted to test into members of the domain and used domain security for the files.
All NT tests were conducted using NTFS (NT File System), and for the Linux tests we used EXT2.
We did a bare minimum of common-sense tuning on each server. For example, both servers had dedicated swap partitions, and we left Samba's parameters set at their defaults.
Oddly enough, Red Hat's default configuration of Samba differs from the documentation's defaults. For NT, we set it to prioritise file sharing and to give no performance boost to foreground applications.
In our tests we wanted to compare the NOSs when serving as full-range file and print servers. Our results were somewhat surprising, and can probably be partly attributed to the fact that we printed directly to the server being tested, rather than to a separate, dedicated print server.
Based on our observation of the server, NT would have been more comparable to Red Hat if it did not have to handle printing duties in addition to file serving. Where Red Hat swallowed print jobs as fast as they came, the queue grew longer and longer under NT during the course of the test, distracting the server from file sharing.
Finally, with Samba we had some problems with stability and reliability - it was so troublesome that we ended up testing with just 30 clients. The same kinds of problems manifested themselves with NT as well, although to a lesser degree and mostly to do with printing.www.redhat.com