On Tue, 22 Sep 2009 18:25:32 +0200 Joachim Schipper <joac...@joachimschipper.nl> wrote:
> On Tue, Sep 22, 2009 at 08:24:43AM +0200, Benoit Chesneau wrote: > > On Mon, Sep 21, 2009 at 12:47 PM, Joachim Schipper > > <joac...@joachimschipper.nl> wrote: > > > > > The "well-known" solution is to use a database for all state and > > > replicate that. A simpler solution is to just use a static site. What > > > issue are you trying to solve? > > > > > > > > > Wan't to replicate a full couchdb (http://couchb.org) node. For now > > replication of couchdb works only per db and over http. So if i have > > 1000 db I will have 1000 http connections open on the same machines. > > It's somehow dangerous. That's why I'm/was looking for a way to > > replicate on the fs level rather than db level. I've since developped > > (yesterday morning) a global change handler that allow me to know in > > quasi rt changes in all dbs of a couchdb node. That could hep for a > > global replication I guess. I will do more tests/try today, but any > > other idee is welcome :) > > I don't think having a thousand databases is a good idea, but exactly > how are a thousand HTTP connections bad? Sure, keepalives will consume > some resources, but it's not like having a thousand connections is all > that impressive on modern hardware. > > Note: I'm not saying it's a good idea, just that it'll work. > > Joachim > What is the current state of the iSCSI QLI driver? At one time I considered using an iSCSI and disk mirroring to achieve something like this. Dhu