[ 
https://issues.apache.org/jira/browse/COMDEV-277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16392986#comment-16392986
 ] 

Mayar Mahmoud commented on COMDEV-277:
--------------------------------------

Hi, 

I am Mayar Abd Elaziz, final year computer engineering student at faculty of 
engineering - Alexandria University. I am interested in working on the load 
balancer. I have investigated the two implementations of the load balancer 
(shared data vs sharding) and understood how *schedule()* works but I couldn't 
relate that to one of the known load balancing algorithms(e.g round robin and 
least connected algorithms).

My thoughts are as follows
 # I am not sure if implementing one of those algorithms will improve the 
performance?
 # There was a problem in choosing the invoker when all the invokers are 
overloaded that is solved (I guess) in the sharding implementation by choosing 
a random invoker.

And I am not sure what do you mean by

"Distributed algorithm for performance would be put in place like work 
stealing, prioritization of the activations, and direct connect for streaming 
capabilities."

_Could you help me to get a clearer view of what can be done to improve the 
load balancer?_

> OpenWhisk GSOC: Performance: Improve Load Balancer (Scala)
> ----------------------------------------------------------
>
>                 Key: COMDEV-277
>                 URL: https://issues.apache.org/jira/browse/COMDEV-277
>             Project: Community Development
>          Issue Type: Brainstorming
>            Reporter: Carlos Santana
>            Priority: Major
>              Labels: gsoc, gsoc2018, openwhisk, serverless
>
> *OpenWhisk GSOC: Performance: Improve Load Balancer (Scala)*
> OpenWhisk today is both a scheduler and load balancer located in the 
> Controller, the student with be working on improving the algorithms for this 
> components to improve the performance.
> The student will have multiple options and would need to use a benchmark to 
> measure effectiveness of the changes.
> Distributed algorithm for performance would be put in place like work 
> stealing, prioritization of the activations, and direct connect for streaming 
> capabilities.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@community.apache.org
For additional commands, e-mail: dev-h...@community.apache.org

Reply via email to