[Gllug] On asking and answering GNU/Linux help questions...

Peter Grandi pg_gllug at gllug.to.sabi.co.UK
Wed Oct 19 15:22:29 UTC 2005


>>> On Wed, 19 Oct 2005 12:30:59 +0100, Ben Fitzgerald
>>> <ben_m_f at yahoo.co.uk> said:

[ ... ]

>> [ ... ] Astonishingly :-) using the obvious keywords for a
>> web search returns quite a number of seemingly relevant
>> results (some of them recent, some older), for example:

>> http://www.justfuckinggoogleit.com/

Just this without the context (that however you put below)
results in a rather unfair quote and comment, whose very
unfairness has been exploited by some creeps in other comments.
Thanks not a lot.

Hey, that's just the first line of a whole set of links. Were it
the only link it would beeen still OK, but that it was not
perhaps makes a rather large amount of difference...

What I write is usually exceptionally terse (e-mail is not the
medium for proper discussions of complex subjects, so one has to
make do), so sometimes I try to use devices like this to convey
hints.

Putting in a selected list of relevant links was meant to convey
something like ''OK, just because I have a bit of spare time and
you did state you were a bit lost I can do a bit of spoonfeeding'',
and putting in as the first link that humorous one was meant to
convey ''but hey, I think that you really could have done your own
homework''. The two messages were by choice together.

ben_m_f> I'm sure the information is on the web. The problem was
ben_m_f> that as a tcp tuning novice I did not know which
ben_m_f> keywords to start on.

Uhm. Well, as I said I used the _obvious_ keywords.

Admittedly I have probably more background on these issues
because I occasionally follow the literature, and this gave me
an advantage, specifically that gigabit TCP tuning is/was indeed
a hot controversial topic (I also hinted this very subtly by
quoting "Ethernet" at one point, as several people argue that
Gigabit ''Ethernet'' is not really such).

Then sometimes it looks like that I can word web searches better
than most, as I probably have some understand how search engines
indexes work, and gives me perhaps an unfair advantage.

But still, the query I used was:

  http://WWW.Google.com/search?num=100&as_epq=gigabit+ethernet&as_q=linux+tcp+performance+tuning
    «"gigabit ethernet" linux tcp performance tuning»

and making it a bit more specific by adding «howto OR faq OR
tutorial» returns some different list of possibly more relevant
results.

BTW, it took me a bit to realize that almost all relevant papers
say "gigabit ethernet" and that using "1ghz ethernet" is not as
good.

BTW, as to the general story, the really big thing about Gigabit
''Ethernet'' is that Ethernet frames on it last way too little,
thus jumbo frames (this I knew beforehand). But all this causes
not just controversy, but potentially anomalies, e.g. if
non-Gigabit segments are involves, if throughput matters less
than latency or if latency is high. It's a huge if usually
successful hack.

ben_m_f> Also I needed an answer quickly. If one were to
ben_m_f> disqualify questions that can be answered through the
ben_m_f> internet surely only queries pushing at the edges of
ben_m_f> information in the public domain should be posted?

Bah! The particular aspect of it was that the keywords were
fairly obvious. My non-novice special knowledge was that I knew
beforehand that there is some literature on Gigabit Ethernet
performance issues, and therefore that a search with even
obvious keywords was likely to succeed.

ben_m_f> I know I cannot match that high standard, so
ben_m_f> sorry... ;-) Questions that seem readily apparent to
ben_m_f> you may be harder for me, especially when under time
ben_m_f> pressure.

That's why I have been especially understanding and have
provided some amount of spoonfeeding, even with a bit of a ''you
could have done it'' shaking of the head.

But the main complaint I had with your question was that it was
too vague (you must admit that a questions that begings with «is
slow» may appear so :->).

As to this some of my general principles on help requests online
(specifically mailing lists/newsgroups here).

The main problem is that email/news is a high latency medium for
some people, merely because they are otherwise busy, for example
myself. So I regard messages/articles as mini-memos, which
should as tersely as possible discuss as much as possible (wider
window) to mitigate the effects of the latency. Other people who
have nothing better to do than look at their inbox/newsreader
all the time tend to snipe lots of 3 line vacuosnesses in real
time, as if email/news were IRC.

As to help questions, front-loading the question (and even on
IRC) with some context and using at least semi-precise, terse,
language, helps a lot not just because of the huge rountrip
delay (the ''human'' one), but because computer puzzles are
usually very, very context dependent, and puzzle solving is much
easier usually if one can see at least some of the big picture.

Perhaps you should have asked:

  ''I am getting performance problems in the following
  situation:

    A beefy quad server on RHAS 3 with the usual RedHat heavily
    2.4.21 kernel. with one 64 bit GB eth card "....", an app
    that every second open 300 (serial|parallel) TCP connections
    to that server (on the same switch|....), sending "..."MiB
    of stuff. I see low CPU utilitization, and a tiny percentage
    of socket buffer overflows, and a ping time of around 2ms.

  The performance problems are that I expect (bandwidth
  X|latency Y) and I am getting only (bandwidth X0/latency
  Y0). Links or advice on what to check/do?''

A bit more detail would have helped more...

>> [ ... various links ... ]

ben_m_f> Thanks for the links above. I'll trawl through these
ben_m_f> when I get a chance.

Uhmmmmmmmmmm, thanks for the appreciation, but this is not a
good thing to say. You mean that I spend some time getting those
links for you and you did not even click on them instantly to
see what they are like? Not even on these:

  http://www-didc.lbl.gov/TCP-tuning/linux.html
  http://datatag.web.cern.ch/datatag/howto/tcp.html

that I put at the top as just by looking at them one sees that
they are really short and terse and really really to the point?

I try to do a bit of spoonfeeding, against my best principles,
and it does not even work? :-(

Or perhaps :-).

ben_m_f> Again, many thanks for replying!

Thanks for the appreciation!

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list