Collection Replication

[Luiz Camargo]

Hello guys,

Has somebody ever had to replicate a large collection (three columns with 5.000 facts, for example ) between nodes (ports)?

The channel collection and the operator (<~) do not to send all facts. They send only about 300 facts. This is a limitation?

How to manipulate large collections on distributed system? 

Thank you 

[Peter Alvaro]

Hi Luiz,

can you say a little more about your use case and expectations?  I admit that I am surprised to hear that so few records are being successfully transmitted!  but our programming model assume that communication isn't reliable.  if you want to reliably transmit information over an unreliable medium, you should use a reliable delivery protocol.

the BUD prototype also probably won't perform very well locally over large collections...

Peter

--
You received this message because you are subscribed to the Google Groups "Bloom Language" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bloom-lang+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Luiz Camargo]

Hi Peter,

I am using the Bloom in an approach for model transformation (MT) (Data-centric MT, PhD research). The large models (1.000.000 elements, for example) are partitioned in collections. I need to distribute them between nodes (ports). In my experiments (using a single machine), I use a master node (port 12345) in which  the slave nodes (ip_port) are connected. Summarizing, I would to processing the model portions in parallel.

Indeed, I have not used the reliable delivery protocol.

I created a protocol:

channel :connect, [:@addr, :slave]

channel :mdl_cast, [:@slave, :addr_count] => [:mdl_name, :mdl_ref, :mdl_type]  

Even knowing that bloom may not perform well, but their characteristics are relevant.

Thank Peter for your attention

To unsubscribe from this group and stop receiving emails from it, send an email to bloom-lang+...@googlegroups.com.