Slashdot Mirror
NT Beats Linux in Round 2
strat writes "PC Week ran it's own benchmarks, with Mindcraft, Microsoft and
Red Hat. The margins were tighter this time, but NT still fared
better. They specifically mentioned the lack of multithreading
in the Linux IP stack as the main bottleneck. I wonder how
4.4lite would have fared? "
Ok, the results are in. The benchmarks, when administered properly, show that NT outperforms Linux in every category.
We now need to prove the open source model instead of confusing the issue by nursing our bruised egos. I'm already seeing the "denial" posts that hark back to OS/2 days. "Well, they may have gotten better numbers in this, but that doesn't really matter." Yadda, yadda, yadda. What is wrong with you people? Can't you accept the results? Quit playing games and start making progress.
Some would love to argue with me until they are blue in the face about how this doesn't mean NT's better, etc. Fine. That is NOT what I'm talking about. The benchmark has exposed an architectural flaw/oversight. We need to fix this in order to reach the performance numbers we need to be a server operating system. So let's do it. Let's fix it. Start downloading and start coding.
I have linux installed on every one of my file servers. I will be able to fend off the criticism for now. I have faith in out community and in our talent. However, if we continue pointless arguements and pulpit pounding about how linux really is better, instead of making sure that it IS better, we're sunk!
And if you can't code, you can always test. Get involved. This is your operating system.
"Doubt your doubts and believe your beliefs."
"Doubt your doubts and believe your beliefs." -- Switchfoot, Ode to Chin
http://cs.alfred.edu/~lansdoct/mstest.ht ml
Kinda puts the whole shebang into perspective. Watch out for the twist of irony at the end.
Regards,
Regards,
-scott
When people choose to dump their MS OS and go with Linux, I've never heard "because Linux serves more pages per second than NT" given as a reason.
The reasons are usually one of the following:
Better Stability
Tired of being locked into MS-only solutions
No license fees
Doesn't need an expensive hardware upgrade
No annoying features like animated paperclips
Linux doesn't pretend to be smarter than you, and then fail at it.
No registry which causes more pain and grief than it actually solves.
It doesn't really matter than NT beats Linux in this test. Sure we should fix the problems, but NT is still NT with all the liabilities which it has become famous for.
Of all the comments I've ever posted, this is definately one of them
http://www.kegel.com/mindcraft_redux.html
(Found at LWN.) This page is full of technical analysis of the Linux kernel and Apache, explaining a number of performance problems that the benchmarks brought to light, as well as solutions for many of them.
It sounds like a repeat test in another month or two would bring things even closer. With khttpd, we might even win.
One of the most important points why more and more companies choose Linux/Apache
has nothing to do with speed or price of the OS or the webserver. The really
expensive part when running webservices are administrative costs. NT & IIS may win here
for *very* small sites, and UNIX wins when it comes to housing more than a few dozen
sites - enough to make NT and especially IIS unmanageable.
* NT has no useful scripting, Linux has everything you can ever need
* you cannot remote administer NT (Im not talking about fast connections here,
(where you could use VNC), try to administer NT over a modem line. Good luck)
* once youve made your decision to use IIS, youre completly stuck when it comes
to changing to another type of webserver, or sometimes even when you want to
transfer sites from on IIS to another. Microsoft has a tool to do that job, but -
guess what - it crashes on even the smallest problems. Apaches configuration
files can easily converted to another text-based config - use sed, awk or whatever
youre use experienced with.
* if something goes wrong with IIS, the event log will contain such useful error
messages "could not bind instance XXX. The data is the error code. 43 00 00 6c".
* IIS is a hell when it comes to logging. All logging is done asynchronously, so your
only chance to see whats going on is to wait a few minutes for IIS to sync() the
logs. Really a pain when you want to study the logs...
* Sometimes under NT, the MMC console simply is stuck. Then your only chance to
get it running again is to restart the system, simply logging in as a different
user does not help. Very annoying.
* Any finally, those beloved situations where those windows popup:
"Your system is running low on virtual memory...". When you check the taskmanager,
it will show you that neither applications nor the system itself seems to use
that much memory. Again, your only chance is to restart the system.
++ Say to Elrond "Hello.".
Elrond says "No.". Elrond gives you some lunch.
Here's what the second benchmark showed:
:^)
1. The Linux Advocates Were Right
Mindcraft *did* seriously mis-configure Linux in the first test.
The Linux/Apache peak performance in the second test was approximately 50% higher than in the first test.
More important, the disasterous collapse of Linux/Apache at higher loads, that occurred in MindCraft's first test, was nowhere to be seen in the second test - Linux/Apache performance remained high as the load increased. The performance drop-off in the first test was caused by MindCraft's mis-configuration.
2. The Anti-Linux Zealots Were Wrong
Linux advocates did *not* oppose the first test simply because NT beat Linux. The opposition was based on valid concerns about how the test was run - concerns that have been born out by the second test.
There is little serious opposition to the second test, which is generally considered fair (within the limits of the benchmark). In fact, the knowledge gained from the second test has been welcomed by the Linux community, who look forward to the performance gains that will result.
3. NT/IIS Beat Linux/Apache - Not That It Matters
IIS on NT *did* achieve a higher benchmark result than Apache (or Samba) on Linux. But, as many have pointed out, the conditions of the benchmark are highly artificial. In the real world, where there is a greater mix of activity on the server, Linux's virtues in the areas of stability, task management, and I/O performance would play a greater role.
As some have pointed out, when you're shopping for a reliable delivery van, the fact that it can be beat by a dragster is of little consequence. Or, to use another car analogy, by pouring on the nitrous, you can beat any other car on the track - for one lap (before burning out your engine).
4. Linux/Apache Performance Was Excellent - Not That It Matters
As others have pointed out, the Linux/Apache performance on one CPU was enough to handle the load of *twenty* T1 lines. But again, the test is too artificial for that to have much meaning.
5. That MindCraft Guy is a Whiner
There are always jerks on every side of every issue. For him to pick out the most obnoxious things said by some Linux supporters, and suggest that they mean something, is childish. Those mouth-offs don't represent the Linux community, anymore than Ballmer represents the...NT...oh...wait...
we had this story, a couple days ago: here... it has the exact same url. this is the same story. doh!