December 20th, 2005

Desktop Linux versus Windows XP shootout

Posted by George Ou @ 12:59 pm

Categories: Infrastructure

Tags:

In my last blog "Could Desktop Linux really be so slow?" I posted some interesting performance numbers of how a mainstream desktop Linux distribution like Linspire performed on a $100 PC.  The performance results for Linspire Linux were quite a disappointment, so I decided to investigate further and ask you, the reader, what Linux distribution would represent desktop Linux better.  Those who criticize Microsoft for producing some of the world’s biggest bloatware clearly don’t have the facts on their side.While I got a lot of helpful suggestions and appreciate all of them, I’m sorry to say that I can’t use all of them or test every single Linux distribution that was recommended.  I’ve decided that the best candidate for testing Linux desktop performance against Windows XP Professional would be SUSE Linux (the free OpenSUSE version), since it is a mainstream distribution and preload optimized in its default configuration.  Having tested it over the weekend, it appears that SUSE is much faster than Linspire and it impressed me enough that I intend to use it in a dual-boot configuration with Windows on my own PC from now on.

One other suggestion was that 128 MBs of RAM simply wasn’t enough to run Linux and KDE even though Windows XP will run at acceptable levels with 128 MBs of RAM.  There are those who will argue with my statement that Windows XP runs properly on 128 MBs of RAM and point out that their Windows XP system runs like molasses with 512 MBs of RAM.  I say to those people that they have a very poorly configured Windows XP system and that it doesn’t reflect what a clean Windows XP system is capable of.  My other reasons for testing the 128 MB configuration is that I believe that any operating system should at least properly boot up and load a single productivity application at a time without choking in the process.  This low-memory torture test assures me that there will be plenty of memory left for my data on a typical PC with 256 to 512 MBs of memory.  One other great point that a reader brought up was that these systems with integrated video cards use a unified memory architecture where system memory is borrowed for non-existent video memory.  It turned out that 32 MBs of RAM was in fact being borrowed for the integrated video adapter such that the system was only left with 96 MBs of memory.  Even so, I still say this is workable for Windows XP Professional so it’s only fair to test desktop Linux in the same hardware configuration.  But in the interest of being thorough, I’ve decided to add a 256 MB DDR-333 (typically $20-$30) stick of memory into the test bed to test the 352-MB RAM configuration in addition to the 96-MB RAM configuration.

Since most people will require the use of some kind anti-virus protection for their Windows PC, I added PC-Cillin Internet Security Suite 2006 to the Windows test bed.  Since this would be a handicap for the Windows platform and not a pure comparison between Windows to Linux OS, I included the results of Windows without anti-virus.  In fairness, desktop Linux would also require anti-virus if it ever reached mainstream status with more than a 20% market share on the desktop, but I’ll have to leave that debate for another day so we can get on with the results.

Software specifications:

Hardware specifications:

• AMD Sempron 2400+ (1667 MHz Socket A) CPU
• SIS motherboard chipset
• 96 MBs DDR-333 RAM (Trial A)
• 352 MBs DDR-333 RAM (Trial B)
• 32 MBs of DDR-333 RAM allocated for integrated video
• SIS on-board integrated video adapter
• Ultra ATA-133 40 GB hard drive
• 52x CD-ROM drive

Results:

As we can see from these OS boot time and application load time results, Microsoft handily beat the open source platform.  What’s more surprising is that Windows XP and Office 2003 running in 96 MBs of RAM even managed to beat OpenSUSE and OpenOffice.org 2 running in 352 MBs of RAM!  While I admit that load times aren’t the only factor for judging performance, it is a significant one.  Any software developer will tell you that load times don’t just magically shrink; it takes careful design and plenty of hard work to optimize code to be responsive.  There are those who have and will always criticize Microsoft for producing some of the world’s biggest bloatware, but they clearly don’t have the facts on their side.  There may be plenty of things to criticize Microsoft for, but bloated software isn’t one of them.