Re: Make GNUN more user-friendly

[Thérèse Godefroy]

Le lundi 02 septembre 2013 à 10:05 +0400, Ineiev a écrit :
> On 08/30/2013 07:15 PM, Thérèse Godefroy wrote:
> > Le vendredi 30 août 2013 à 15:07 +0000, Ineiev a écrit :
> >> On 08/30/2013 12:35 PM, Thérèse Godefroy wrote:
> >>> Le vendredi 30 août 2013 à 16:13 +0400, Ineiev a écrit :
> >>>> I'm not sure it wouldn't be done faster.
> >>> 2 seconds for each regeneration + the time it takes to write the command
> >>> line once:
> >>> ~/GNU/www-fr $ ./batch-transvalidate -p philosophy/stallman-kth.fr.po
> >>> Then another second to refresh the browser page.
> >> Of course it must be faster to run GNUN locally.
> > 
> > That's the whole point.  :)
> 
> Nonetheless I'll try to check whether we can run it on a server.

Then GNUN will become a SASS!   lol

Might be a good idea, though, because people wouldn't have to worry
about installing it. But the modified PO files would have to be uploaded
the "easy" way, meaning via the browser, not CVS, otherwise no non-geek
translator would ever use the facility. 

Besides, one has to take the connection speed into account. How long
would it take for a person living in a remote French village, with a
6km-long, rotten ADSL subscriber line (not my case, don't worry), to
upload gpl-faq.fr.po (385.6kB) and get the HTML back? Certainly more
than 3 seconds, I would think.

And what about the recurrent overload of some backbones in the Boston
area?

Here is what I saw on Nov. 29, 2012, around 19:00 UTC, using
speedtest.net:

              DL (Mb/s)               UL (Mb/s)
Boston        0,92 and 1.27           0.85
New York      8.82                    0.85
Seattle       1.67 and 3.95           0.82

and, by comparison, on Dec. 1, 2012, around 8:30 UTC:

              DL (Mb/s)               UL (Mb/s)
Boston        10.70 and 10.09         0.82
New York      9.61 and 10.66          0.85
Seattle       8.68 and 6.90           0.84
[In that case, my own ADSL connection may have been limiting.]

I once saw a 1900ms ping to Savannah but 220ms seems very frequent
(whereas the ping to New-York is 120ms). What may be more relevant, tho,
is what I recorded with Wireshark on Nov. 29 and Dec. 1, 2012 during cvs
checkouts of www, more or less simultaneously with the speed tests. It
took 1h 18min on Dec 1 and 5min 30s on Dec 1 to checkout 381MB. What
stands out is the significant packet loss on Nov 29, which compounds
with the lower download speed. The "conversations" of Nov. 29 and Dec. 1
are summarized in the attached screenshots. You can see that more
packets and more bytes were exchanged on Nov. 29, than Dec.1, and that
the overall download speed was consistent with the speedtests. So, in
that case, the rate-limiting component was not the CVS server but the
network itself. 

BTW, traceroute showed that I was connected via AboveNet on both Nov. 19
and Dec. 1. They seem to live up to their bad reputation.

Thérèse

wireshark-conversations-cvs.png
Description: PNG image