> On 03/10/2014 05:53, Kouhei Kaigai wrote:
> > Yep, that's my pain. Even though usual query does not take many
> > buffers pinned, my use case needs to fetch megabytes scale data at
> > once because of performance reason; page-by-page synchronous scan makes
> GPU being idle.
> Doesn't your GPU have an async queue and exec mechanism? Then you could
> do an asyn DMA to the GPU with an event, use that event in he GPU to start
> the kernel and in the DB to release the pin?
>
That is exactly what I'm doing now. Problem is, it needs to keep multiple
buffers (likely, more than a few dozen thousands) in the queue not to make
GPU waiting for the data to be calculated. Thus, I want to pin relatively
many buffers than usual query workloads.
Thanks,
--
NEC OSS Promotion Center / PG-Strom Project
KaiGai Kohei <kaigai@ak.jp.nec.com>