Slashdot Mirror
OSS Web Stacks Outperformed by .Net?
Gimble writes "eWeek has an article up that looks at the performance of portals using open source stacks and comparing them to their MS equivalents. The article's conclusion is that .Net outperforms the open source stacks, mainly because of its tighter integration, but also notes that running the open source stacks on Windows (WAMP) delivered strong performance." From the article: "Based on our forays into user forums for many top open-source enterprise applications, there are many IT managers attempting to run open-source products on Windows servers--attracted, no doubt, to the benefits and efficiencies of using open source without having to become Linux administrators. The results of our WAMP stack tests indicate that these folks might be on to something."
But this is a very close second place.
#1. NO tuning was done on the LAMP stuff. None at all. They ran the stuff "out of the box".
#2. They didn't write their own app. That means they didn't test the SAME processes on each system.
#3. They didn't bother to find WHERE the differences were. Is it in the IP stack? Is it in the OS? Is it in the scripting language? Is it in the app?
How bad can "research" be and still be published in "eWeek"? There wasn't any research done for that article.
Microsoft has, in the past, taken various short-cuts when IIS was the server and IE was the browser. Is that the case in this "study"? Are the other "stacks" "slower" because they follow the protocols?
You won't know because they'd didn't LOOK for the REASON behind their "results".
At least MindCraft was paid to do poor research.
They seem to have a generalized poverty of data. Their charts seem absurd to the point of being straw men. I mean, come on - I don't think there's anything seriously wrong enough with Linux that WAMP would have a score of 12 transactions/sec, competing with Windows, whereas LAMP would have a performance of 2. My experience with Windows vs. Linux has always been that they are similar in terms of speed from pure processing tasks to 3d games. Sometimes Windows does a little better, sometimes Linux is better. But they're usually in the same ballpark. The numbers are just too neat. It's like they put up a chart saying that Republicans, Germans, Koreans and Canadians have sex once a month, whereas Democrats, Brazilians, and the British have sex five million times per second.
Moreover, the whole rest of the article is morass of poetic circumlocution. My gut feeling as somebody who works with words a lot is that they're trying to obfuscate something with a giant wall of banal text. I don't know exactly what that is, because I don't feel like reading all of it, but if I had to guess I'd say that the real thing to take away from this article is that anybody can set up .NET and a Windows box, but that it requires a little bit of patience and research to make Linux work properly - research that these people were not willing to do.
Here's the price breakdown for a SINGLE webserver that allows external connections to authenticate (non-domain, say a e-commerce site with user accounts) against a SINGLE SQL 2005 Database. Sql Express is free, however it's not licensed for unlimited users in a production environment.
Web Server (Prices from CDW.com):
- 1 Copy Windows 2003 R2 (5 CALs): $959.99
- 1 External Connector License for Windows: $1,969.99
- Total: $2,929.98
SQL Server:- 1 Copy Windows 2003 R2 (5 CALs): $959.99
- 1 Copy SQL Server 2005 Workgroup Edition 1 Processor License: $3,819.99
- Total: $4,779.98
Grand Total for a single web and sql box: $7,709.96.And don't forget that you'll need SOME hardware to run that OS. Even barebones boxes with no data protection will run you $500 a box.
So, to start a basic e-commerce site on the legit, you're talking roughly $9,000 for windows and $1,000 for Linux/OSS.
TOUGH sell for Microsoft for the little guy.
You raise some excellent points. If I may be allowed to expand upon them ...
#1. Set a price limit. You can set multiple limits ($1,000 vs $5,000 vs $25,000 vs $100,000 vs $1,000,000+). The key concept here is that you get different characteristics as your budget increases/decreases. What characteristics does each "stack" offer in each price range? Yep, this does give the advantage to Free stacks (Free like speech, free like beer). Deal with it. In the Real World the bottom line is the bottom line. Each team gets to spend the money however they want to.
#2. Get the "experts" to tune each stack. BUT they must document each modification they make, including WHY they made that modification (what testing did they run and how did those test results tell them what mod's to make) AND they are only allowed to make mod's that can be found via public websites (no secret tuning parameters that are only known to the organization writing that software) AND they aren't allowed to touch any source code. They get what everyone else gets.
#3. The fun part. Each team gets to pull apart the work of the other teams. Even if your solution is faster for the specs given, how much wiggle room do you have? Is faster and fragile better than slower and stable? How much "slower" is acceptable for how much more "stable"? Can the other team defeat your security (network access only)?
#4. Freeze those systems. Then, over the next year, patch them and re-test them. Do the patches break the "tuning" that was done?
Now that would be an informative test process (and would result in lots of articles and interviews for the magazine publishing it).
Yeah, you can run Linux / Apache / MySQL / perl on a single drive workstation and get damn good performance for less than $300.
But that will be completely different from Oracle / Java on a cluster of Suns costing $10,000,000. And not just in the number of boxes you'd be running.
Now, I know that Plone is a dog, and XOOPS may be popular on Sourceforge, but I don't think it's the most obvious choice for building a high performance portal using PHP. So, using these two as the basis for testing is silly.
The fact that JBoss Portal on Windows outpaces JBoss Portal on Linux has a simple reason: JBoss isn't fully open source; crucial parts of it (namely the Java runtime itself) are under Sun's control, and hell will freeze over before Sun bothers to do a good job implementing Java for their competitors' Linux systems.
As for things generally running faster on Windows, that's implausible. Differences between raw Windows and Linux system performance are at most in the single digit percentages, so if they saw any significant differences between the same applications running on top of the two platforms, either the application vendor spends more time tuning for Windows (as in Sun Java), or the testing labs screwed up.
In fact, the whole test is really ill conceived: none of the "portals" they compared provide the same functionality; it just doesn't make sense to test them against each other. Overall, this test mostly seems to test the competency of eWeek, and they aren't doing too well.