Thanks for the reply. I've discussed NT performance problems with
people who've looked into the visible nooks and crannies. They tell
me that
a) interrupt latency on NT is about two orders of magnitude slower
than Linux. This is due to the model they implement called IORequests
IIRC.
b) System calls are very expensive on NT, though I don't have the
numbers. This is what caused them to move graphics drivers into the
kernel for NT4. Their original design wasn't working. They went to
great lengths with the original graphics subsystems, using a kernel
thread rendevous and shared memory to help alleviate the syscall
bottleneck. It worked better than straight calls, but not well
enough.
I heard some real numbers from a collegue at (an unnamed) company
using NT to serve Web pages. They bought 60 Dual Proc PII/450's with
an appropriate amount of memory to handle their Web serving load which
was mostly small banners and cookies. They found that each machine
would 'fall over' about once a week and they'd have to reboot it. The
obvious symptom was that the CPU could not respond to the NumLock key
in real time. My collegue did a test of their load by diverting a
copy of the incoming requests to a GNU/Linux box. He measured
performance and found that two of those machines would handle the load
with cycles to spare. Albeit this is annecdotal, yet I don't think
there is anything unique about this experience.
-O
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/