> Edric M Ellis <email@example.com> wrote in message <firstname.lastname@example.org>... >> "Sujit Das" <SujitRDas@gmail.com> writes: >> > Thanks Edric. That is very insightful. Yes, we realize that data >> > serialization does happen during transfer. We have also encountered >> > the limit of 2 GB ( I believe) in 64-bit workers during serialization. >> >> If you have 64-bit workers and a 64-bit client, the limit no longer >> applies (it was removed in R2013a). > > That is great news. Does the overall performance of worker-to-client > data transfer improve with this new version? I would read up the > release notes for R2013a.
The performance didn't change when the restriction was removed. Do you have a simple example of what you're trying to do where you think the performance needs to be improved?
> On a different note, we also experience slowness in closing matlabpool > especially when we run 100s of workers. What exactly matlabpool does > other than destroying temp files/dirs and closing connections to > workers after the workers are completed?
That's basically it. Do you have some figures for how long it takes against number of workers? What cluster type are you using?