#3149: Data.HashTable is slow
------------------------------+---------------------------------------------
Reporter: dons | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: libraries/base | Version: 6.10.1
Severity: normal | Resolution:
Keywords: | Testcase:
Os: Unknown/Multiple | Architecture: Unknown/Multiple
------------------------------+---------------------------------------------
Comment (by dons):
GC destroys this benchmark (and Data.HashTable in general, it seems).
Removing lists from the equation:
{{{
import Control.Monad
import qualified Data.HashTable as H
n = 10000000
main = do
m <- H.new (==) H.hashInt
forM_ [1..10000000] $ \n -> H.insert m n n
v <- H.lookup m 100
print v
}}}
51s by default, 98% GC:
{{{
$ time ./A
Just 100
./A +RTS 48.83s user 1.18s system 98% cpu 50.929 total
1,787,760,112 bytes allocated in the heap
3,333,907,256 bytes copied during GC
836,979,232 bytes maximum residency (11 sample(s))
4,261,176 bytes maximum slop
1628 MB total memory in use (25 MB lost due to fragmentation)
Generation 0: 3400 collections, 0 parallel, 44.14s, 45.15s elapsed
Generation 1: 11 collections, 0 parallel, 3.78s, 4.96s elapsed
INIT time 0.00s ( 0.00s elapsed)
MUT time 0.52s ( 0.52s elapsed)
GC time 47.91s ( 50.11s elapsed)
EXIT time 0.00s ( 0.02s elapsed)
Total time 48.44s ( 50.63s elapsed)
***%GC time 98.9% (99.0% elapsed)***
Alloc rate 3,416,378,481 bytes per MUT second
Productivity 1.1% of total user, 1.0% of total elapsed
}}}
Turning up the heap hint:
{{{
$ time ./A +RTS -H1G
Just 100
./A +RTS -H1G 6.68s user 0.99s system 98% cpu 7.778 total
1,787,825,320 bytes allocated in the heap
2,021,056,256 bytes copied during GC
466,523,872 bytes maximum residency (2 sample(s))
5,748,312 bytes maximum slop
1117 MB total memory in use (17 MB lost due to fragmentation)
Generation 0: 9 collections, 0 parallel, 4.68s, 5.03s elapsed
Generation 1: 2 collections, 0 parallel, 1.16s, 1.23s elapsed
INIT time 0.00s ( 0.00s elapsed)
MUT time 0.84s ( 1.49s elapsed)
GC time 5.84s ( 6.26s elapsed)
EXIT time 0.00s ( 0.17s elapsed)
Total time 6.69s ( 7.75s elapsed)
***%GC time 87.4% (80.8% elapsed)***
Alloc rate 2,120,087,562 bytes per MUT second
Productivity 12.6% of total user, 10.9% of total elapsed
./A +RTS -sstderr -H1G 6.69s user 0.90s system 96% cpu 7.834 total
}}}
And GC is stil 89% of the time. If we could remove the GC as a factor we'd
be looking at something like 0.6s on this benchmark, instead of 50s.
------------------------------
Using the parallel GC is fun, since you get really serious GC use:
{{{
$ ghc -O2 A.hs --make -threaded -no-recomp
on the commandline:
Warning: -no-recomp is deprecated: Use -fforce-recomp instead
[1 of 1] Compiling Main ( A.hs, A.o )
Linking A ...
$ time ./A +RTS -H1G -sstderr -N2
./A +RTS -H1G -sstderr -N2
Just 100
1,787,833,280 bytes allocated in the heap
2,477,128,208 bytes copied during GC
672,983,168 bytes maximum residency (3 sample(s))
44,186,984 bytes maximum slop
1349 MB total memory in use (21 MB lost due to fragmentation)
Generation 0: 14 collections, 14 parallel, 7.63s, 4.17s elapsed
Generation 1: 3 collections, 2 parallel, 3.47s, 2.12s elapsed
...
Parallel GC work balance: 1.89 (309610432 / 164021678, ideal 2)
}}}
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/3149#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
