On Sat, Mar 30, 2013 at 2:09 PM, Roy Smith <[email protected]> wrote: > In article <[email protected]>, > Chris Angelico <[email protected]> wrote: > >> Especially facepalm because there's some way to do this that's faster >> than straight INSERT statements, and it's not clearly documented as >> "hey, guys, if you want to dump loads of data in, use COPY instead" >> (it might be that, I don't know, but usually COPY isn't directly >> transliterable with INSERT). > > We're actually using REPLACE INTO. For the initial data load, we could > just as well do INSERT, but we need the REPLACE functionality as we roll > in the daily incremental updates. > > This also explains why, even after provisioning our RDS instance for > 2000 IOPS (that's AWS-speak for "we paid extra to get more disk > bandwidth"), we didn't see any performance improvement!
Hmm. I heard around the forums that Amazon weren't that great at disk bandwidth anyway, and that provisioning IO was often a waste of money. But we never did all that much much research on Amazon I/O performance; shortly after doing some basic benchmarking, we decided that the cloud was a poor fit for our system model, and went looking at dedicated servers with their own RAID storage right there on the bus. ChrisA -- http://mail.python.org/mailman/listinfo/python-list
