[tor-dev] Tor metrics: Questionable relays-by-platform accounting

[Karsten Loesing]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 08/01/16 16:54, Fabian Keil wrote:
> Karsten Loesing <karsten at torproject.org> wrote:
> 
>> On 07/01/16 18:48, Fabian Keil wrote:
>>> Looking at "Tor metrics - Relays by platform" [0] one could
>>> think that the number of "FreeBSD" relays is only slightly
>>> lower than the number of "Windows" relays.
>>> 
>>> It also looks like the number of "other" platforms is
>>> constantly zero while the number of "Darwin" relays is unclear
>>> (but potentially zero as well).
>>> 
>>> Obviously that's incorrect:
>>> 
>>> fk at r500 ~ $sudo -u _tor grep platform 
>>> /usr/jails/tor-jail/var/db/tor/cached-descriptors | cut -w -f 5
>>> | sort | uniq -c | sort -rn 6662 Linux 555 Windows 286 FreeBSD
>>> 83 OpenBSD 25 Darwin 10 NetBSD 4 SunOS 4 ElectroBSD 4 Bitrig 2 
>>> DragonFly 1 GNU/kFreeBSD 1 CYGWIN_NT-10.0-WOW
>>> 
>>> Is it possible that the "Tor metrics" count everything but
>>> "Linux" and "Windows" as FreeBSD?
>> 
>> Good question.  Not exactly: it's counting everything containing
>> "BSD" as "FreeBSD".  Ugh.  I agree that this is misleading.  How
>> about we change the graph legend to "BSD"?
> 
> Without further explanation that seems still a bit misleading
> because "DragonFly" and "Bitrig" are BSDs, too. In fact,
> "DragonFly" is short for "DragonFly BSD" so even a graph legend
> like "platforms that contain 'BSD'" would be misleading for someone
> who does not know the actual platform strings.

Well, it still seems like an improvement over the current situation
where we're wrongly classifying non-FreeBSD-BSDs as FreeBSD.

Oh, and "platforms that contain 'BSD' wouldn't fit in the legend anyway.

Should I proceed with my earlier suggestion to change "FreeBSD" to
"BSD" in the legend?

>> (Unfortunately, we cannot distinguish easily between the other
>> BSDs in that graph, because we'd have to re-import the archives
>> since 2007.)
> 
> That's indeed unfortunate.

Only so many battles to fight...

>> Note that your query above is not entirely correct, either.
>> You're overcounting relays publishing more than one descriptor,
>> whereas Metrics only counts descriptors being referenced from the
>> consensus. The latest numbers from Tor metrics are:
>> 
>> 2016-01-05,,,,Darwin,,22, 2016-01-05,,,,FreeBSD,,371, 
>> 2016-01-05,,,,Linux,,6384, 2016-01-05,,,,Other,,9, 
>> 2016-01-05,,,,Windows,,387,
>> 
>> See https://metrics.torproject.org/servers-data.html
> 
> I see, thanks.

Sure.

All the best,
Karsten


> Fabian
> 
> 
> 
> _______________________________________________ tor-dev mailing
> list tor-dev at lists.torproject.org 
> https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-dev
> 

-----BEGIN PGP SIGNATURE-----
Comment: GPGTools - http://gpgtools.org

iQEcBAEBAgAGBQJWj99oAAoJEJD5dJfVqbCrkjcIAJIEBNFMv3yjXLpTVmpcOKdi
yIUTItds05GgVDN2Wdb5NbvtJe5HDEZfZMPPHLLcOEYmZejhkoQ7K2Nur5fA+Wcm
bhtjkDsBdSCl7aa54DADp1FH8YGTzUJ0fPHqZt6VkAZWAoWE5C8Rqj+2cE4ztCTp
C6rLSXh+YH7LOh1RVhUSiOVUG9AA/77TMx6N9o09soW8Rk7G5kA4ryL8VyENZGhe
ddMp20Os6HMqmCG5b/j5l7fozWdwtCAmXg94sP/x1hrxzp+aN6NtaEi+y7CU5c0M
0IrRgH/NS4dkkC215IMYuzfzB4iDFYkAL537b/kwp4X23aPxfjOpY5NMkVYM7+o=
=jCBT
-----END PGP SIGNATURE-----