Comments on: Clustered JavaSpaces, good or bad? http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2006%2F01%2F03%2Fclustered-javaspaces-good-or-bad%2F%23comment-&seed_title=Clustered+JavaSpaces%2C+good+or+bad%3F Mon, 02 Jan 2012 08:22:24 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Scattered Thoughts » Blog Archive » Reliable Applications from Unreliable Components http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2006%2F01%2F03%2Fclustered-javaspaces-good-or-bad%2F%23comment-4&seed_title=Clustered+JavaSpaces%2C+good+or+bad%3F/comment-page-1/#comment-4 Scattered Thoughts » Blog Archive » Reliable Applications from Unreliable Components Mon, 13 Feb 2006 16:51:27 +0000 http://www.dancres.org/blitzblog/?p=8#comment-4 [...] As I’ve said in a previous posting, I’m not a fan of infrastructure level clustering. It basically comes down to the fact that this kind of approach to resilience and scale is achieved through centralization and strict control of the environment. Whilst such a level of control and centralization might be possible in certain well-defined, small scale circumstances, it gets much more difficult across a network of any size and with any more than a few machines. [...] [...] As I’ve said in a previous posting, I’m not a fan of infrastructure level clustering. It basically comes down to the fact that this kind of approach to resilience and scale is achieved through centralization and strict control of the environment. Whilst such a level of control and centralization might be possible in certain well-defined, small scale circumstances, it gets much more difficult across a network of any size and with any more than a few machines. [...]

]]> By: Dan Creswell http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2006%2F01%2F03%2Fclustered-javaspaces-good-or-bad%2F%23comment-3&seed_title=Clustered+JavaSpaces%2C+good+or+bad%3F/comment-page-1/#comment-3 Dan Creswell Tue, 10 Jan 2006 08:51:01 +0000 http://www.dancres.org/blitzblog/?p=8#comment-3 You are making sense and that's one approach though not the one I would adopt. There are a number of ways to handle replication/state-loss (publicly documented on the web and elsewhere) which can be used to solve this problem. I have constructed several different solutions which I'm currently productizing. Thus, due to commercial constraints, I'm not able to provide further details at this point. JINI/JavaSpaces does indeed help avoid SPOF's but you potentially need other things as well, in particular appropriate algorithms/architecture. I can only offer the following: If you think about the solution in terms of clustering or other solutions you're familiar with, you're unlikely to find the other options. You are making sense and that’s one approach though not the one I would adopt.

There are a number of ways to handle replication/state-loss (publicly documented on the web and elsewhere) which can be used to solve this problem. I have constructed several different solutions which I’m currently productizing. Thus, due to commercial constraints, I’m not able to provide further details at this point.

JINI/JavaSpaces does indeed help avoid SPOF’s but you potentially need other things as well, in particular appropriate algorithms/architecture. I can only offer the following:

If you think about the solution in terms of clustering or other solutions you’re familiar with, you’re unlikely to find the other options.

]]> By: Lachlan O'Dea http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2006%2F01%2F03%2Fclustered-javaspaces-good-or-bad%2F%23comment-2&seed_title=Clustered+JavaSpaces%2C+good+or+bad%3F/comment-page-1/#comment-2 Lachlan O'Dea Tue, 10 Jan 2006 02:04:16 +0000 http://www.dancres.org/blitzblog/?p=8#comment-2 I'm pretty new to this stuff. I can understand the significant problems inherent with clustering, but I'm wondering what alternative approaches exist for using Javaspaces. If I correctly understand how Blitz works currently, a Javaspace instance can be hosted only on one machine. This then becomes a single point of failure, which was one of the problems I was hoping Jini/Javaspaces would help me avoid. So what is a good approach to avoid a single point of failure, without using clustering? The only idea I have is to use multiple Javaspaces, with one or more workers asynchronously copying objects between them as needed. Clients somehow discover the Javaspaces that are available and manually failover between them. Am I making sense at all? I’m pretty new to this stuff. I can understand the significant problems inherent with clustering, but I’m wondering what alternative approaches exist for using Javaspaces. If I correctly understand how Blitz works currently, a Javaspace instance can be hosted only on one machine. This then becomes a single point of failure, which was one of the problems I was hoping Jini/Javaspaces would help me avoid.

So what is a good approach to avoid a single point of failure, without using clustering? The only idea I have is to use multiple Javaspaces, with one or more workers asynchronously copying objects between them as needed. Clients somehow discover the Javaspaces that are available and manually failover between them. Am I making sense at all?

]]>