Graham,If one of the slaves goes down in the time until the master finds out the slave will keep getting requests even so it isn't available. It is also difficult to reliably keep lots of machines in sync so that they all return the same data, especially when there are lots of different admins for the system, each admining their own slave inconsistently with the other slave.
If it was an easy task all 3 xapi servers would have an high uptime.
Shaun On 9 Sep 2009, at 08:08, grahamjones...@googlemail.com wrote:
Tom,I am surprised that you think reliability is the issue - I thought we could make it quite reliable by having multiple redundant servers. I expected the criticism to be complexity - to make it work you need something along the lines of the following: 1. A 'master' which maintains a database of the available servers, whether they are alive or not, and how fast they are. 2. Some 'deputies' which synchronise themselves with the master and take over its duty if the master dies (the 'take over' bit is the bit I don't know how to do - will have to do something with DNS I suppose...). 3. When a request comes in the current master decides which sever to send the request to and passes it on.Therefore I think the main issue is that this is rather complicated and may be difficult to maintain, rather than a reliability issue as such? There is of course the additional processing step of choosing a server, which could slow things down, but I suppose it is just a scan through the bounding boxes of the various servers to see which ones can meet this request, so it might not add much onto the overall request.Graham. On Sep 8, 2009 8:46pm, Tom Hughes <t...@compton.nu> wrote: > On 08/09/09 19:41, Graham Jones wrote: > > > >> This would work for a single 'main' server, but I like the idea of it>> being distributed with lots of little ones (for example the computer in> > my attic could serve Northern England, someone else could do Belgium >> etc.). I don't know how to deal with re-directing the requests without> > a central main server though...any ideas? > > > >> That kind of distribution is, in my opinion, a terrible idea. Such a system will just never be reliable.> > > > Tom > > > > -- > > Tom Hughes (t...@compton.nu) > > http://www.compton.nu/ >_______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev
