Runtime comparisons

[kbruegge]

This is what it looks like so far. The makefile performs measurements and produces the plots (if you have all the python stuff installed 😉 )

Storm is not really worth testing at the moment because it become horribly slow for some reason.

[kbruegge]

I added another xml. I only enqueue the important events

        <!--Only every 1000nth event will be put into the queue-->
        <streams.DataRate every="1000" logmemory="true"/>
        <Enqueue condition="%{data.@datarate} != null" queue="passDataRate"/>

[alexeyegorov]

This is what I got with the lastest XML on my macbook using one core (important: also with streams-storm)

[kbruegge]

Some things we need to fix I think:

• At the moment command line flags are not propagated to the xml it seems. Did you get that working?
• The one xml file should work for all backends.
• The storm backend shoudl return errocode 0 when it has processed all items.

[alexeyegorov]

I just submitted the most current streams-storm version, which stops the spout with error code 0.

[alexeyegorov]

At the moment, there is a problem with queues in streams-storm. I am on this, but at least using the output attribute combined with a Skip operator can replace a simple queue with Enqueue. Maybe using this combination for the performance measurement?

[alexeyegorov]

streams-storm should now quit the application after all data items have been processed (there was a bug that is now fixed; although, Jenkins seems to be unreachable and the package will be pushed to maven repo ASAP).

[kbruegge]

Quite awesome. This works for me now. Calling make automatically compiles, runs and plots everything without problems. I changed the makefile back to renaming the output using a mv command. I think we can merge that now. The parameter stuff should still be fixed at some point though. What do you think?.

[alexeyegorov]

Gread. Yeah, I need to still fix the parameter stuff. I will open up an issue here and work on it (storm and flink at least).