Avery Ching commented on GIRAPH-41:

I certainly agree that the implementation of all graph mutations and messages 
should be done through the same message passing mechanism between workers.  
Graph mutation requests are nothing more than a specific message.  In fact, 
aggregators should be allowed to aggregate all graph mutations as well as 

> Change graph mutation RPC API to a kind of message
> --------------------------------------------------
>                 Key: GIRAPH-41
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-41
>             Project: Giraph
>          Issue Type: Improvement
>          Components: bsp
>            Reporter: Hyunsik Choi
> Graph mutation functions (e.g., addVertexRequest, addEdgeRequest..) directly 
> invoke RPC functions. 
> In processing, these RPC calls may incur delays caused by TCP round-trip time 
> and communication overheads caused by frequent RPC call. Especially, when 
> many workers try to mutate vertices and edges simultaneously, the 
> synchronization overheads may also occur in receiving sides. It may become 
> intensive as the size of cluster increases.
> If we change graph mutation API to a kind of messages, it would be more 
> efficient. If possible, graph mutation message API and data messages API 
> (i.e., sendMsg)can be integrated into one message passing API.
> What do you think about that?

This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira


Reply via email to