openoffice-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From janI <j...@apache.org>
Subject Re: wiki, forum, www performance upgrade.
Date Sat, 06 Apr 2013 16:37:30 GMT
On 6 April 2013 13:37, Rob Weir <robweir@apache.org> wrote:

> On Sat, Apr 6, 2013 at 5:51 AM, janI <jani@apache.org> wrote:
>
> > Hi.
> >
> > Now that our wiki (mwiki) runs smoothly and performs quite well, it is
> > slowly time to take a look at our other services.
> >
> > In order to determine what should/can be done, I need some figures.
> >
> > Can anybody help me with traffic figures for
> >    wiki.o.o, forum.o.o (total for the vm not per language) and www.o.o
> >
> > Does end-users download directly from www.o.o or from a secondary site ?
> >
> > I need the following numbers (if possible):
> >    - daily number of clicks (average)
> >    - peak number of clicks within a short period like 1 hour
> >
> >
>
> From Google Analytics we have:
>
> For wiki:  around 33K page views/day
>
> For forums, I have no idea.  They are using a Google Analytics account (
> UA-1087265-8) that is not connected with the rest of the website.  If that
> could be updated to match the rest of the website we'd have typical numbers
> in a week.
>

Thanks for your answers, but I am a bit scared you (as a PMC) tell me
"they" use an account unknown to us. Who has the forum account ?

I have the technical power to change it (sudo access) but if a PMC tell we
do no know who owns it, it would not be right just to change it. Who but
the "owner" can authorize a change ?

I trust this is a misunderstanding somewhere, it cannot be that we run a
service, that we do not control, in the sense of change management.


> For the website as a whole (all subdomains tracked) it is around 730K/day
> for a peak day.  If you subtract out the forum traffic and the downloads
> (which serviced by SourceForge) then the static website is probably more
> like 600K/day visits.
>

Why do I need to subtract forum, I thought they ran as a separate dns ? do
we not count on basis of our template (that loads the GA account).


>
> So that indirectly answers your other question -- the big download files
> are from the SourceForge distribution network.  But we do have smaller
> downloads, of documentation PDF's.  These are mainly from the wiki.
>
> In terms of peaks, our biggest day usually Wednesday, lowest on Sunday.
> The variation there around 200K/visits.
>

Super just what I need.


>
> Hourly peak is around 15K visits for all tracked subdomains together.
>

Do you know the division between mwiki, forum and www ?

>
> Note:  for all of these I'm talking about "pages", where a single page
> could include several HTTP calls, for the HTML, the Javascript, images,
> etc.  Since anyone with Javascript not enabled would not register, the
> actual numbers are probably a little bit higher than this.
>

Good point, that is very important for a a traffic server, where e.g.
javascripts would be cached.

I am trying to convince Infra, that we should have a common trafficserver
for wiki and forum (and maybe www). Currently we use eresbus as proxy, and
all packets are forwarded to the ooo-wiki2-vm where that trafficserver
filters cached pages. Forum goes direct to httpd, so every packet causes
mysql activity. If we had a trafficserver in eresbus, we would for wiki2
save loads of network traffic and for forum network traffic as well as
mysql activity.


>
> In any case, one approach would be to enable GA for the Forums, track that
> for a week or two and then compare that to the Wiki numbers.  If the stacks
> are similar, then we could think of the traffic ratio between the two and
> tune based on that.
>

I agree, but not being PMC, I do not feel I am in the position to "just"
make that change. Who "owns" the forum, they should authorize such a
change. I can make the change (provided you tell give me the GA line to
insert) when I am sure I am not doing a change that someone else depend on.


>
>
> > I would like to tune at least forum before we release 4.0 to prepare for
> a
> > higher load. Tuning of www depend on a discussion with infra, because it
> is
> > a shared service.
> >
> >
> If it is like the release of 3.4.0 and 3.4.1 we'll get a few hot pages:
>
> 1) The announcement blog post.  With both 3.4.0 and 3.4.1 releases these
> were heavily linked to by the press.  Roller went down in the 3.4.1 case.
> I don't know if that was due to the traffic or not.  But we might want to
> plan on having the announcement post be a static webpage just to be sure.
>

The blog post is on the general apache server and not our www, is that
correct ?

>
> 2) The homepage and http://www.openoffice.org/download page will be hot.
> You can see from the download chart the boost we get when a new release
> occurs:  http://www.openoffice.org/stats/downloads.html  (AOO 3.4.0 was
> May
> 8th 2012 and AOO 3.4.1 was Aug 23rd).
>

That is a real obvious case for trafficserver, and until it is in place I
can ask to have the httpd use memcached (which it might be already).
Currently the www is on a general apache server, so I need strong arguments
to make changes (high load is a strong argument)


>
>
>
>
> > Translate-vm also needs tunings, but that is in progress, and infra has
> > been very kind and prepared an extra vm, so we/I install/tune without
> > affecting the active system.
> >
> > thx in advance for figures.
> > rgds
> > jan I.
>

THANKS a lot for your answer, which help me a lot doing calc.. Sorry for
being frank about the forum, but it really make me nervous that we have a
service where part it unknown.

rgds
Jan I

> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message