whole lotta kays

whole lotta kays

[Steinar Bang]

Just an observation, from the fact that rgnus 0.82 is counting the
kilobytes, as it's transferring the headers:
 There can be quite a few k's of data transferred for not *all* that
 many unread messages.  I've seen 100k for some 100-odd messages, in
 some cases

Variations are presumably caused by ticked articles, and the setting
of nntp-nov-gap (by experimenting, I found long lived groups, with
lots of articles, to open a *lot* faster, when decreasing this value).


- Steinar

Re: whole lotta kays

[Lars Magne Ingebrigtsen]

Steinar Bang <sb@metis.no> writes:

> Variations are presumably caused by ticked articles, and the setting
> of nntp-nov-gap (by experimenting, I found long lived groups, with
> lots of articles, to open a *lot* faster, when decreasing this value).

What seems to be an optimal value for that variable?  

-- 
(domestic pets only, the antidote for overdose, milk.)
  larsi@ifi.uio.no * Lars Ingebrigtsen

Re: whole lotta kays

[Steinar Bang]

>>>>> Lars Magne Ingebrigtsen <larsi@ifi.uio.no>:

> Steinar Bang <sb@metis.no> writes:
>> Variations are presumably caused by ticked articles, and the setting
>> of nntp-nov-gap (by experimenting, I found long lived groups, with
>> lots of articles, to open a *lot* faster, when decreasing this value).

> What seems to be an optimal value for that variable?  

I did some tests on some no.* groups last autumn, but I didn't save
them anywhere.

I ended up having nntp-nov-gap set to 1, which made these groups load
a lot faster, but probably slowed down groups with lots of new
articles. 

Then I "solved" the problem by creating nnml-folders where I could
move articles I wanted to save from these groups, deleted (or whatever
the Gnus term was again) all the ticked articles, and used the default
for nntp-nov-gap.

Would a more formalized test, with timings for groups with different
characteristics, and different settings of nntp-nov-gap, be of
interest?

If so: what numbers should I primarily look at?


- Steinar

Re: whole lotta kays

[Lars Magne Ingebrigtsen]

Steinar Bang <sb@metis.no> writes:

> Would a more formalized test, with timings for groups with different
> characteristics, and different settings of nntp-nov-gap, be of
> interest?

It would indeed.  The current default is just something I picked
without testing at all.

> If so: what numbers should I primarily look at?

The length of time it takes to enter the groups?  :-)

Well, varying `nntp-nov-gap' over a range from 1 to 40 to nil, and
trying to select groups that have lots of articles and few articles,
and selecting groups that have lots of read articles in between the
unread articles.

-- 
(domestic pets only, the antidote for overdose, milk.)
  larsi@ifi.uio.no * Lars Ingebrigtsen

some nntp-nov-gap timings

[Steinar Bang]

>>>>> Lars Magne Ingebrigtsen <larsi@ifi.uio.no>:

> Steinar Bang <sb@metis.no> writes:
>> Would a more formalized test, with timings for groups with different
>> characteristics, and different settings of nntp-nov-gap, be of
>> interest?

> It would indeed.  The current default is just something I picked
> without testing at all.

>> If so: what numbers should I primarily look at?

> The length of time it takes to enter the groups?  :-)

I meant: "which function should I look at in elp-results"

> Well, varying `nntp-nov-gap' over a range from 1 to 40 to nil, and
> trying to select groups that have lots of articles and few articles,
> and selecting groups that have lots of read articles in between the
> unread articles.

Well you're not getting that, because doing these variations manually
just takes too much time.  But I got the following numbers today, when
accessing the foreign group
	nntp+news.eunet.no:no.alt.motorsykler
over a 64kb line.  The numbers are the value of nntp-nov-gap, followed
by the time used by gnus-group-read-group, with an average of 5 tries.
The group at this point, had 3 unread articles, and 120 ticked
articles. 

nntp-nov-gap	average time used (sec.)
1		20.80
5		21.09
10		33.23
20		55.16
30		63.16
40		74.73

I have no idea of the load on the NNTP server, or the general network
traffic at the time of measurement.  I guess that a more fair
measurement, would involve more averages, with the different values of
nntp-nov-gap interleaved.

Maybe something for some emacs lisp guru to hack together?


- Steinar

Re: some nntp-nov-gap timings

[Lars Magne Ingebrigtsen]

Steinar Bang <sb@metis.no> writes:

> nntp-nov-gap	average time used (sec.)
> 1		20.80
> 5		21.09
> 10		33.23

I've now changed the default to 5.

-- 
(domestic pets only, the antidote for overdose, milk.)
  larsi@gnus.org * Lars Magne Ingebrigtsen