experience with very large numbers of nodes?

[Miles Fidelman <mf...@meetinghouse.net>]

Hi Folks,

Does anybody have any experience with very large Couch deployments - on 
the order of 100s to 1000s of nodes?

I'm looking at a peer-to-peer document management application, where 
every user has a copy of Couch on their desktop or a nearby LAN server.

I know that Couch's replication model purports to scale to very large 
numbers of nodes - but has anybody actually tried it, and lived to tell 
the tale?

Thanks,

Miles Fidelman

-- 
In theory, there is no difference between theory and practice.
In<fnord>  practice, there is.   .... Yogi Berra

Re: experience with very large numbers of nodes?

[Miles Fidelman <mf...@meetinghouse.net>]

Eric Casteleijn wrote:
> On 04/20/2010 01:17 PM, Miles Fidelman wrote:
>> Hi Folks,
>>
>> Does anybody have any experience with very large Couch deployments - on
>> the order of 100s to 1000s of nodes?
>
> For Ubuntu One, we have a very large number of nodes: one on every 
> machine of every user of the service, and then a central node on our 
> server (or set of nodes, but each user will only see one) that it 
> replicates from and to, and we believe this can be made to scale quite 
> well, although we are still in the tuning phase.
Sounds like a good datapoint.  Thanks!
>
> Each user has their own set of databases though, so there is way less 
> replication needed and chance of conflicts, than if you were to have 
> one global database that thousands of nodes all replicate between 
> eachother.
We're in a situation where we're supporting small-to-mid-sized groups, 
each with their own set of documents (think workgroups in a large 
corporation).  A particular database would be replicated across anywhere 
from 10 to 100 machines - individual laptops and desktops, plus 
workgroup servers.  I'm thinking that each user would synchronize with a 
local workgroup server, and the workgroup servers would synchronize with 
each other.

Miles

-- 
In theory, there is no difference between theory and practice.
In<fnord>  practice, there is.   .... Yogi Berra

Re: experience with very large numbers of nodes?

[Eric Casteleijn <er...@canonical.com>]

On 04/20/2010 01:17 PM, Miles Fidelman wrote:
> Hi Folks,
>
> Does anybody have any experience with very large Couch deployments - on
> the order of 100s to 1000s of nodes?
>
> I'm looking at a peer-to-peer document management application, where
> every user has a copy of Couch on their desktop or a nearby LAN server.
>
> I know that Couch's replication model purports to scale to very large
> numbers of nodes - but has anybody actually tried it, and lived to tell
> the tale?

For Ubuntu One, we have a very large number of nodes: one on every 
machine of every user of the service, and then a central node on our 
server (or set of nodes, but each user will only see one) that it 
replicates from and to, and we believe this can be made to scale quite 
well, although we are still in the tuning phase.

Each user has their own set of databases though, so there is way less 
replication needed and chance of conflicts, than if you were to have one 
global database that thousands of nodes all replicate between eachother. 
(In which case I would probably still choose a star topology with a 
central server that everyone replicates with, rather than going with 
true p2p where everyone replicates with everyone, but that's my 
intuition and not a carefully thought out objection. There may be ways 
to make that work.)


-- 
eric casteleijn
https://code.launchpad.net/~thisfred
Canonical Ltd.