About compilation time

Eyal Rozenberg E.Rozenberg at cwi.nl
Thu Dec 10 18:09:48 CET 2015



On 09-12-15 08:54, Stefan Manegold wrote:
>> In addition, looking at the results, compile time is order of tens of ms. Is
>> this what I should expect?
>
> Sound reasonable to me.

An additional note from my experience...

Mustafa, I would guess you're configuring all of the DBMSes in your 
comparison to utilize the CPU as much as possible, right? With MonetDB, 
that would specifically mean setting your DB's nthreads value to match 
what your hardware supports. So, in my experience the compilation time 
changes, often significantly, for different values of nthreads.

In my old group at Hua Wei research, we actually added timing hooks for 
measurement, and a few of these are visible here:
http://is.gd/Ktd74q
bottom-right chart (note all columns are normalized to 100%; the 
brownish-dark-red blocks at the bottom are MonetDB's query 
comiplation/optimization, with all columns named 'AXE' being nthread=1 
for reasons I won't go into). We observed a ratio of over 2:1 between 
32-thread and single-thread query compilation. We also observed some 
non-negligible variance in timing.

Caveat: This was with a version of MonetDB from 2013, so things may have 
changed.


More information about the developers-list mailing list