Thanks for your comments. As suggested I ran a small benchmark of both
versions. Turns out that the difference between (ref-set ref @ref) and
ensure is huge ...
(defn write-skew [de-ref]
(let [cats (ref 1)
dogs (ref 1)
john (future
(dosync
(when (< (+ @cats (de-ref dogs)) 3)
(alter cats inc))))
mary (future
(dosync
(when (< (+ (de-ref cats) @dogs) 3)
(alter dogs inc))))]
(do @john @mary)
(dosync
(> (+ @cats @dogs) 3))))
user> (bench/with-progress-reporting
(bench/bench (write-skew (fn [x] (ref-set x @x))) :verbose))
Warming up for JIT optimisations 10000000000 ...
compilation occurred before 18580 iterations
Estimating execution count ...
Sampling ...
Final GC...
Checking GC...
Finding outliers ...
Bootstrapping ...
Checking outlier significance
amd64 Linux 3.16.0-4-amd64 4 cpu(s)
OpenJDK 64-Bit Server VM 24.121-b00
Runtime arguments: -Dfile.encoding=UTF-8 -XX:+TieredCompilation
-XX:TieredStopAtLevel=1 -XX:-OmitStackTraceInFastThrow
-Dclojure.compile.path=/home/bertschi/GitRepos/Vorlesungen/PS2/target/classes
-Dps2.version=0.0.0 -Dclojure.debug=false
Evaluation count : 1675560 in 60 samples of 27926 calls.
Execution time sample mean : 37,327714 µs
Execution time mean : 37,325021 µs
Execution time sample std-deviation : 1,137506 µs
Execution time std-deviation : 1,147482 µs
Execution time lower quantile : 34,971387 µs ( 2,5%)
Execution time upper quantile : 39,511899 µs (97,5%)
Overhead used : 12,902114 ns
nil
user> (bench/with-progress-reporting
(bench/bench (write-skew ensure)))
Warming up for JIT optimisations 10000000000 ...
compilation occurred before 40 iterations
compilation occurred before 61 iterations
Estimating execution count ...
Sampling ...
Final GC...
Checking GC...
Finding outliers ...
Bootstrapping ...
Checking outlier significance
Evaluation count : 60 in 60 samples of 1 calls.
Execution time mean : 1,662454 sec
Execution time std-deviation : 1,906919 sec
Execution time lower quantile : 186,394000 µs ( 2,5%)
Execution time upper quantile : 7,630998 sec (97,5%)
Overhead used : 12,902114 ns
Found 4 outliers in 60 samples (6,6667 %)
low-severe 2 (3,3333 %)
low-mild 2 (3,3333 %)
Variance from outliers : 98,3142 % Variance is severely inflated by
outliers
nil
I'm running clojure 1.8.0 by the way. According to VisualVM the (ref-set
ref @ref) version spends most of its time in
clojure.lang.LockingTransaction$RetryEx.<init> () while ensure gets
basically stuck (> 99%) at clojure.lang.LockingTransaction.tryWriteLock ().
Am Mittwoch, 8. März 2017 14:40:01 UTC+1 schrieb Herwig Hochleitner:
>
> 2017-03-06 12:06 GMT+01:00 'bertschi' via Clojure <clo...@googlegroups.com
> <javascript:>>:
>
> > From the docs it says "Allows for more concurrency than (ref-set ref
> @ref)".
> > I would read that as "runs at least as fast as (ref-set ref @ref)", but
> > using (ref-set dogs @dogs) instead of (ensure dogs) and the same with
> cats,
> > does not exhibit these occasional runtime spikes.
> >
> > Am I misunderstanding something or is it a bug in ensure?
>
> If ensure does indeed show worse (or not even better) performance than
> (ref-set ref @ref) in benchmarks, it certainly would go against my
> understanding of clojure's STM and I'd consider this a bug (since
> ensure could just be implemented by ref-set for a free performance
> boost, in this case).
>
> To show this behavior and move this along into a ticket, it would be
> helpful to run those two examples in a microbenchmark runner, like
> criterium and attach a profiler to uncover locking problems and
> excessive retries, that would likely be the root cause.
>
--
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to clojure@googlegroups.com
Note that posts from new members are moderated - please be patient with your
first post.
To unsubscribe from this group, send email to
clojure+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
---
You received this message because you are subscribed to the Google Groups
"Clojure" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to clojure+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
