Linked by Thom Holwerda on Fri 9th Sep 2011 15:17 UTC
Windows More news on Windows 8. This time around, Gabe Aul, a director of program management in Windows, blogged about the changes Microsoft has made to Windows 8's boot process. The results are impressive - a boot time not much slower than waking from sleep on current Windows 7 and Mac OS X machines. This is, of course, a vital component of getting Windows NT ready for tablets.
Thread beginning with comment 489275
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE[4]: They always promise this
by zlynx on Tue 13th Sep 2011 00:23 UTC in reply to "RE[3]: They always promise this"
Member since:

I'd like to address this point specifically, because you greatly misunderstand me. I read on LWN a Linux kernel developer talking about **LINUX**. In his opinion Linux has a bug if it cannot cope better with "A bunch of IO heavy tasks started simultaneously" than it can with "The same tasks started in sequence." He was saying that Linux should do better than the user, always, in scheduling IO or it's a bug in Linux. He never mentioned Windows.

What **I** did was begin thinking about the differences in the way I observed Linux and Windows reacting to disk contention, which lead me to run some experiments, which lead me to the conclusion that I began with above: Windows sucks at scheduling IO.

Somewhat related to this is my personal observations of Linux and Windows I/O behavior on a laptop with only 1 GB of RAM back in 2006.

Surprisingly, Windows XP seemed to perform much better under serious load than Linux 2.6.24. The main problem that Linux had is managing read activity under severe memory pressure. Reading data in always required writing data out to swap and/or releasing data from disk cache.

Linux almost always got this wrong, writing out data to swap or releasing buffer cache that was soon required, while Windows XP seemed to be able to keep a relevant working set in RAM for each application.

Linux also created absolutely horrid I/O patterns, writing a few KB to swap, then seeking back to read a few KB from a file, then seeking back to write a few more KB to swap, then needing to read a bit from other swap, etc. On a laptop drive this decreased I/O throughput to less than 50 KB/second while Windows XP seemed to be able to swap out a few megabytes at a time and then read entire files into cache. This may also have a lot to do with how Windows prioritizes the foreground application.

I sure hope the Linux developers have managed to fix the awful VM behavior. I know there have been a lot of fixes made to it.

I probably won't see the differences since I no longer have any 1 GB machines.

Reply Parent Score: 2

sorpigal Member since:

Your experiences are not out of line with mine. I did stop using swap on Linux years ago on my desktop systems, or rather set swappiness so that it was only ever used in dire emergencies. This does make make things remarkably better. And, as you note, Windows uses swap often without as bad an effect.

It is my understanding that there's a lot of tuning you can do if you want to try to make this sort of thing better, but I haven't looked in to it.

Reply Parent Score: 2