[NTLUG:Discuss] Internet connection speed tests - XP vs Linux

Chris Cox cjcox at acm.org
Wed Jul 11 11:25:19 CDT 2007


terry wrote:
> I've done a number of speed tests (testing Internet connection [bandwidth]),
> using 2 PCs; MS Windows XP system and a Linux system, (same speed test on
> each).  It appears that 4 out of 5 times, the Linux system usually gets
> faster speed test results.  I'm using the latest Firefox browser on both
> systems and going to the same bandwidth test site on each and running them
> one right after the other.
> (Both systems are fully updated - latest patches etc.)
> I did about 4 or 5 tests a few days ago and while I did not calculate the
> exact percentages, I did see fairly consistent faster speeds from the Linux
> system.
> 
> This time, I calculated and Linux was faster on all but two tests;
> Here are results [percentages]:
> Test#                download (%)               upload(%)
>   1                       6    [faster]                   50[faster]
>   2                       32  [faster]                   55[faster]
>   3                       11 [slower]                   55 [faster]
>   4                       21 [faster]                    0  [about the same]
>   5                       15 [faster]                    62 [slower]
> Does anyone have any explanation for this?
> Has anyone else done speed test comparisons like these?
> 
> Is it possible that Linux has a more efficient TCP stack?
>    Or....?

Well... first off, there's a LOT of variables.  So, the test you did,
while it is "real world", it's not scientific... so you can really
make any definitive statement about the results.

In other words... you really haven't proved or shown anything that
I'd draw any conclusions from.

AFAIK, Microsoft's TCP/IP stack is pretty much their own thing but that
they did use the BSD stack for assistance.  Obviously Linux programmers
desire their stack to be good... and I'd say they did a good job.
But your data has too many variables to draw any conclusion even about
the relative network performance between Linux and Windows.

Personally, I'm a bit surprised at the results (which I don't trust).
Numbers might be different, but not with that much of a delta (I would
think).



More information about the Discuss mailing list