Re: Hierarchical fanout with Beam combiners?

2023-05-30 Thread Robert Bradshaw via dev
On Tue, May 30, 2023 at 10:37 AM Kenneth Knowles wrote: > > On Sat, May 27, 2023 at 4:20 PM Stephan Hoyer via dev > wrote: > >> On Fri, May 26, 2023 at 2:59 PM Robert Bradshaw >> wrote: >> >>> Yes, with_hot_key_fanout only performs a single level of fanout. I don't >>> think fanning out more th

Re: Hierarchical fanout with Beam combiners?

2023-05-30 Thread Kenneth Knowles
On Sat, May 27, 2023 at 4:20 PM Stephan Hoyer via dev wrote: > On Fri, May 26, 2023 at 2:59 PM Robert Bradshaw > wrote: > >> Yes, with_hot_key_fanout only performs a single level of fanout. I don't >> think fanning out more than this has been explored, but I would imagine >> that for most cases

Re: Hierarchical fanout with Beam combiners?

2023-05-27 Thread Stephan Hoyer via dev
On Fri, May 26, 2023 at 2:59 PM Robert Bradshaw wrote: > Yes, with_hot_key_fanout only performs a single level of fanout. I don't > think fanning out more than this has been explored, but I would imagine > that for most cases the increased IO would negate most if not all of the > benefits. > My

Re: Hierarchical fanout with Beam combiners?

2023-05-26 Thread Robert Bradshaw via dev
Yes, with_hot_key_fanout only performs a single level of fanout. I don't think fanning out more than this has been explored, but I would imagine that for most cases the increased IO would negate most if not all of the benefits. In particular, note that we already do "combiner lifting" to do as muc

Hierarchical fanout with Beam combiners?

2023-05-26 Thread Stephan Hoyer via dev
We have some use-cases where we are combining over very large sets (e.g., computing the average of 1e5 to 1e6 elements, corresponding to hourly weather observations over the past 50 years). "with_hot_key_fanout" seems to be rather essential for performing these calculations, but as far as I can te