Am 08/01/2011 09:07 PM, schrieb Rob Weir:
On Mon, Aug 1, 2011 at 1:20 PM, Marcus (OOo)<marcus.m...@wtnet.de> wrote:
Hi,
AFAIK the current projects at Apache doesn't have high download numbers
compared with OOo. So, a download request can point directly to a mirror or
a mirror list is shown and the users have to choose themselves from where to
get the software best.
For OpenOffice this doesn't work. With download numbers up to 500,000 per
month we need a more flexible and scaleable solution. I don't know if or how
fast we could reach these numbers again. But even with the half it would be
a lot to handle. Furthermore we have to present a) a simple way to download
for the end-users that is b) *not* stressing a few/specific mirror servers
to their maximum.
The current solution you can see on "download.openoffice.org" works very
well. However, the infrastructure behind this is still hosted on Oracle
server. So, we have to transfer this, too.
I don't speak about domain, website and its content. That's already on the
list for migration. It's about the download redirector behind the colored
webpages (http://download.services.openoffice.org/files/). Here MirrorBrain
is used (www.mirrorbrain.org) to recognize from where the download request
comes and which mirror server to choose that is near to the user.
I would like to continue this service and IMHO it shouldn't be that
difficult when I look at the requirements:
http://www.mirrorbrain.org/requirements/
The obvious question would be: Is there a problem with using the 262
mirrors that Apache already uses?
http://www.apache.org/mirrors/
Do you think that they cannot handle the load?
That's not the question. Otherwise I could present this list:
http://distribution.openoffice.org/mirrors/#mirrors
;-)
No, the real question is not the load but how to do the load balancing
and how to present the best download link to the user.
From what I'm reading, the Apache system does not just give a static
list of mirrors to the user. There is some logic to ensure that the
same server is not at the top all the time. And there are mechanisms,
like closer.cgi to suggest a reasonable mirror for a given user, e.g.:
http://www.apache.org/dyn/closer.cgi
Maybe it's a pre-chosen thing to present a download link. But still it's
a list of servers the user has to choose on there own. And only after
they have seen that only the first link is the pre-chosen one and the
others are nice to know. ;-) For me it's not clear enough.
I wonder whether it would make sense to try the Apache mirroring
infrastructure until we're convinced that it won't work?
Depends on what you define here with "working". Technically it will
work. Maybe Apache will ? a bit because of the suddenly increased load
(when we switch the download method from one to the other day). But the
user hasn't a good solution when searching for a download. See my answer
to Ross.
When running in a VM it should be a small part within the OpenOffice project
to maintain. In the past at Oracle we just had to add/modify/delete mirrors
and very few times to restart the VM in case of problems.
Marcus
Marcus
