From: Vincent Cunniffe (vincent at domain beacon.ie)
Date: Wed 19 May 1999 - 15:59:59 IST
>While I was thinking about why on earth M$ would want to give Linus a million
>dollars, I started thinking about Mindcraft. I read the article the other day
>on slashdot and previous articles a few weeks ago. Whats the general take on
>this survey? I can't believe that NT is decent enough to outperform both Linux
>and Solaris in high volume situations. Is this just a case of Redmond blowing
>crap out their ass again? My own experience with NTServer is that it needs a
>good rebooting often and that it doesn't take a lot to make it fall over.
>Thats on our little network with 8 machines and 4 regular users. We run the
>webserver off it but that doesn't get many hits. I'd hate to be running
>something the size of Yahoo or AOL on it - actually I'd be just too scared!
>I'd have to live in work pe4rmanently and ditch any social life I have. Is the
>survey accurate and why are the main figures in the Linux community so slow to
>respond to its findings?
Having read much of the comments on various sites, including responses and
comments from people such as Linus and Alan Cox, the general feeling is that
the entire survey is designed from the ground up to be as hostile to Linux as
possible.
The hardware and task in question seem to have been specifically chosen to
make Linux perform badly, and the software on the Linux side (kernel, Samba
and webserver) were not tuned effectively at all. The RAID driver in question
was carefully chosen, with a full release quality multithreaded driver for NT,
and an alpha-quality single-threaded driver for linux. There are several
hardware RAID controllers (Mylex, for example, IIRC) which have high-quality
and reliable drivers for Linux.
The task is serving static webpages and files : using a single high-power
server for this, as someone pointed out, is lunacy. It's a single point of failure,
it's extremely expensive, and it cannot be upgraded. Someone went to the
trouble of designing an 8 box, twin processor farm which was cheaper,
a lot more upgradeable, and vastly more reliable in terms of redundancy,
and which was much more suitable for the job in question.
In terms of pure hardware, NT probably does run faster than Linux on the
box in question (4*Xeon), because it was specifically tuned to do so.
Linux at present is more comfortable with lower-end (i.e. twin P-III class)
machines, due to the more granular nature of the Linux 2.2 kernel. The
2.3.x series of kernels is already underway, however, with some serious
work being done on kernel granularity, which should significantly improve
the performance on high-end machines.
It's also worth noting that NT will *not* run effectively on anything more
powerful : it will be faster, but it starts seriously losing marginal gains on
anything above 4 processors.
The Mindcraft survey is probably best worth noting for the amount of
seriously negative press that it generated for Micro$oft : there was a
surprisingly large amount of antagonistic response from normally
non-technical sources such as business magazines and Salon, all of
which carefully explained how the survey was unfair, and which examined
who paid for it and structure by which non-useful results could be canned
and NDA'ed by the company commissioning the survey.
With regard to Solaris, it consistently outperforms NT on everything, as
far as I know.
Vin
This archive was generated by hypermail 2.1.6 : Thu 06 Feb 2003 - 13:04:13 GMT