On Sun, 13 Aug 2017 10:30:34 +0530
Bhavishya <bhavishyagopesh(a)gmail.com>
wrote:
> Hello,
> So I'm trying to find areas where py3 falls behind py2 in terms of
> performance.Augment reasons and if possible, solutions.
>
> And my mentor(@Botanic) suggested if we could write benchmarks for some
> modules which presently lack explicit bms and thus we tried writing bm for
> "threading" module and another to compare concurrency implemented using
> "threading"(which is …
[View More]actually restricted by gil) and using
> "multiprocessing" across py2 & py3
I'm uncertain that this is a good approach. Measuring the performance
of a single module doesn't necessary tell us anything about the
performance of real-world applications. You first have to assess whethow
the performance of said module impacts real-world applications. This
probably requires some domain-specific knowledge, which either you have
or would have to inquire about.
Regards
Antoine.
[View Less]
Hello,
So I'm trying to find areas where py3 falls behind py2 in terms of
performance.Augment reasons and if possible, solutions.
And my mentor(@Botanic) suggested if we could write benchmarks for some
modules which presently lack explicit bms and thus we tried writing bm for
"threading" module and another to compare concurrency implemented using
"threading"(which is actually restricted by gil) and using
"multiprocessing" across py2 & py3
Thank you
On Aug 12, 2017 2:14 PM, <speed-…
[View More]request(a)python.org> wrote:
> Send Speed mailing list submissions to
> speed(a)python.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://mail.python.org/mailman/listinfo/speed
> or, via email, send a message with subject or body 'help' to
> speed-request(a)python.org
>
> You can reach the person managing the list at
> speed-owner(a)python.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of Speed digest..."
>
>
> Today's Topics:
>
> 1. Re: Threading benchmark (Bhavishya)
> 2. Re: Threading benchmark (Bhavishya)
> 3. Re: Threading benchmark (Antoine Pitrou)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Sat, 12 Aug 2017 05:02:07 +0530
> From: Bhavishya <bhavishyagopesh(a)gmail.com>
> To: Victor Stinner <victor.stinner(a)gmail.com>
> Cc: speed(a)python.org
> Subject: Re: [Speed] Threading benchmark
> Message-ID:
> <CAFRL-_coLaZRrBsNceoPO2QLEHa+wAKDdAooHS--E7i1+pew6w@mail.
> gmail.com>
> Content-Type: text/plain; charset="utf-8"
>
> Sorry but I realised that the actual bottleneck is with "range" (and
> enumeration is slower in py3.)
> https://stackoverflow.com/questions/23453133/is-there-a-
> reason-python-3-enumerates-slower-than-python-2.So
> it's not fair to say that multiprocessing is slow...
>
>
> On Sat, Aug 12, 2017 at 4:37 AM, Bhavishya <bhavishyagopesh(a)gmail.com>
> wrote:
>
> > Hello,
> >
> > WHY-> Presently there are no benchmarks for concurrency,
> >
> > METHOD USED-> This s
> > <https://gist.github.com/bhavishyagopesh/76529e7bf170948b4d3f3a51f6e6bc
> 9a>cript
> > tries to benchmark "Concurrency" implemented using threading and
> > multiprocessing module.Actually "threads" in cpython are restricted by
> > "GIL",so it's not actually concurrent...On the other hand
> "multiprocessing"
> > module creates whole different processes but there is substaintial cost
> > involved in spawing a whole new process.
> > So the there is a trade-off involved which is evident as we increase
> > "CRUNCH_NO" variable.
> >
> > ToNote-> Interestingly Here py2 looks faster.(actually by a considerable
> > margin)
> > I'm attaching graphs that compare py2 and py3.
> >
> > IMPROVEMENTS-> More use cases(apart from number crunching)which could
> > verify this and reproduce similar thing.
> >
> > SO Suggestions required?
> >
> > thank you.
> >
> >
> > On Fri, Aug 11, 2017 at 4:09 AM, Victor Stinner <
> victor.stinner(a)gmail.com>
> > wrote:
> >
> >> I don't understand what you are trying to test. For example, for a
> >> lock, it's very different if a single thread uses the lock, or if two
> >> threads use the lock. None of your benchmarks seem to measure
> >> concurrency.
> >>
> >> Victor
> >>
> >> 2017-08-11 0:33 GMT+02:00 Bhavishya <bhavishyagopesh(a)gmail.com>:
> >> > Hello,
> >> >
> >> > I as a part of my gsoc project(and on suggestion from my mentor )tried
> >> > writing a bencmark for threading module.
> >> > It uses "perf" to measure crearion of objects like
> >> Thread,Semaphore,Locks
> >> > etc. and few methods on them but It might lack on experience and the
> >> > correctness of doing so.
> >> >
> >> > Here's the bm .
> >> >
> >> > And here are the results of a run on my local machine [specs]:
> >> >
> >> >
> >> > python3 bm_threading.py
> >> > .....................
> >> > basic: Mean +- std dev: 14.4 us +- 0.4 us
> >> > .....................
> >> > condition: Mean +- std dev: 18.1 ns +- 1.1 ns
> >> > .....................
> >> > lock: Mean +- std dev: 18.1 ns +- 0.9 ns
> >> > .....................
> >> > rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
> >> > .....................
> >> > semaphore: Mean +- std dev: 298 ns +- 56 ns
> >> > .....................
> >> > timer: Mean +- std dev: 6.74 us +- 0.65 us
> >> >
> >> >
> >> >
> >> > python2 bm_threading.py
> >> > .....................
> >> > basic: Mean +- std dev: 20.5 us +- 1.1 us
> >> > .....................
> >> > condition: Mean +- std dev: 148 ns +- 2 ns
> >> > .....................
> >> > lock: Mean +- std dev: 21.1 ns +- 0.7 ns
> >> > .....................
> >> > rlock: Mean +- std dev: 151 ns +- 3 ns
> >> > .....................
> >> > semaphore: Mean +- std dev: 336 ns +- 4 ns
> >> > .....................
> >> > timer: Mean +- std dev: 12.5 us +- 0.7 us
> >> >
> >> > So suggestions?
> >> >
> >> > Thank You
> >> >
> >> > _______________________________________________
> >> > Speed mailing list
> >> > Speed(a)python.org
> >> > https://mail.python.org/mailman/listinfo/speed
> >> >
> >>
> >
> >
>
[View Less]
On Sat, 12 Aug 2017 06:10:00 +0530
Bhavishya <bhavishyagopesh(a)gmail.com>
wrote:
> I just ran a bm
> <https://gist.github.com/bhavishyagopesh/ff0d9053bcd827aac4a594fec3d30ce4>
> that does this number_crunching task(serially), and results are->
>
[...]
>
> So what're your thoughts?
Before looking at benchmarks it would be nice to know what you are
trying to investigate. What is your GSoC topic and who is your mentor?
Regards
Antoine.
I just ran a bm
<https://gist.github.com/bhavishyagopesh/ff0d9053bcd827aac4a594fec3d30ce4>
that does this number_crunching task(serially), and results are->
py2:
number_crunching: Mean +- std dev: 35.9 ms +- 1.0 ms
py3:
number_crunching: Mean +- std dev: 64.1 ms +- 4.1 ms
And after turning *x to Long *in py2:
number_crunching: Mean +- std dev: 168 ms +- 11 ms
So what're your thoughts?
thank you
On Sat, Aug 12, 2017 at 5:02 AM, Bhavishya <bhavishyagopesh(a)gmail.com>
wrote:…
[View More]
> Sorry but I realised that the actual bottleneck is with "range" (and
> enumeration is slower in py3.) https://stackoverflow.com/
> questions/23453133/is-there-a-reason-python-3-enumerates-
> slower-than-python-2.So it's not fair to say that multiprocessing is
> slow...
>
>
> On Sat, Aug 12, 2017 at 4:37 AM, Bhavishya <bhavishyagopesh(a)gmail.com>
> wrote:
>
>> Hello,
>>
>> WHY-> Presently there are no benchmarks for concurrency,
>>
>> METHOD USED-> This s
>> <https://gist.github.com/bhavishyagopesh/76529e7bf170948b4d3f3a51f6e6bc9a>cript
>> tries to benchmark "Concurrency" implemented using threading and
>> multiprocessing module.Actually "threads" in cpython are restricted by
>> "GIL",so it's not actually concurrent...On the other hand "multiprocessing"
>> module creates whole different processes but there is substaintial cost
>> involved in spawing a whole new process.
>> So the there is a trade-off involved which is evident as we increase
>> "CRUNCH_NO" variable.
>>
>> ToNote-> Interestingly Here py2 looks faster.(actually by a considerable
>> margin)
>> I'm attaching graphs that compare py2 and py3.
>>
>> IMPROVEMENTS-> More use cases(apart from number crunching)which could
>> verify this and reproduce similar thing.
>>
>> SO Suggestions required?
>>
>> thank you.
>>
>>
>> On Fri, Aug 11, 2017 at 4:09 AM, Victor Stinner <victor.stinner(a)gmail.com
>> > wrote:
>>
>>> I don't understand what you are trying to test. For example, for a
>>> lock, it's very different if a single thread uses the lock, or if two
>>> threads use the lock. None of your benchmarks seem to measure
>>> concurrency.
>>>
>>> Victor
>>>
>>> 2017-08-11 0:33 GMT+02:00 Bhavishya <bhavishyagopesh(a)gmail.com>:
>>> > Hello,
>>> >
>>> > I as a part of my gsoc project(and on suggestion from my mentor )tried
>>> > writing a bencmark for threading module.
>>> > It uses "perf" to measure crearion of objects like
>>> Thread,Semaphore,Locks
>>> > etc. and few methods on them but It might lack on experience and the
>>> > correctness of doing so.
>>> >
>>> > Here's the bm .
>>> >
>>> > And here are the results of a run on my local machine [specs]:
>>> >
>>> >
>>> > python3 bm_threading.py
>>> > .....................
>>> > basic: Mean +- std dev: 14.4 us +- 0.4 us
>>> > .....................
>>> > condition: Mean +- std dev: 18.1 ns +- 1.1 ns
>>> > .....................
>>> > lock: Mean +- std dev: 18.1 ns +- 0.9 ns
>>> > .....................
>>> > rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
>>> > .....................
>>> > semaphore: Mean +- std dev: 298 ns +- 56 ns
>>> > .....................
>>> > timer: Mean +- std dev: 6.74 us +- 0.65 us
>>> >
>>> >
>>> >
>>> > python2 bm_threading.py
>>> > .....................
>>> > basic: Mean +- std dev: 20.5 us +- 1.1 us
>>> > .....................
>>> > condition: Mean +- std dev: 148 ns +- 2 ns
>>> > .....................
>>> > lock: Mean +- std dev: 21.1 ns +- 0.7 ns
>>> > .....................
>>> > rlock: Mean +- std dev: 151 ns +- 3 ns
>>> > .....................
>>> > semaphore: Mean +- std dev: 336 ns +- 4 ns
>>> > .....................
>>> > timer: Mean +- std dev: 12.5 us +- 0.7 us
>>> >
>>> > So suggestions?
>>> >
>>> > Thank You
>>> >
>>> > _______________________________________________
>>> > Speed mailing list
>>> > Speed(a)python.org
>>> > https://mail.python.org/mailman/listinfo/speed
>>> >
>>>
>>
>>
>
[View Less]
Sorry but I realised that the actual bottleneck is with "range" (and
enumeration is slower in py3.)
https://stackoverflow.com/questions/23453133/is-there-a-reason-python-3-enu…
it's not fair to say that multiprocessing is slow...
On Sat, Aug 12, 2017 at 4:37 AM, Bhavishya <bhavishyagopesh(a)gmail.com>
wrote:
> Hello,
>
> WHY-> Presently there are no benchmarks for concurrency,
>
> METHOD USED-> This s
> <https://gist.github.com/bhavishyagopesh/…
[View More]76529e7bf170948b4d3f3a51f6e6bc9a>cript
> tries to benchmark "Concurrency" implemented using threading and
> multiprocessing module.Actually "threads" in cpython are restricted by
> "GIL",so it's not actually concurrent...On the other hand "multiprocessing"
> module creates whole different processes but there is substaintial cost
> involved in spawing a whole new process.
> So the there is a trade-off involved which is evident as we increase
> "CRUNCH_NO" variable.
>
> ToNote-> Interestingly Here py2 looks faster.(actually by a considerable
> margin)
> I'm attaching graphs that compare py2 and py3.
>
> IMPROVEMENTS-> More use cases(apart from number crunching)which could
> verify this and reproduce similar thing.
>
> SO Suggestions required?
>
> thank you.
>
>
> On Fri, Aug 11, 2017 at 4:09 AM, Victor Stinner <victor.stinner(a)gmail.com>
> wrote:
>
>> I don't understand what you are trying to test. For example, for a
>> lock, it's very different if a single thread uses the lock, or if two
>> threads use the lock. None of your benchmarks seem to measure
>> concurrency.
>>
>> Victor
>>
>> 2017-08-11 0:33 GMT+02:00 Bhavishya <bhavishyagopesh(a)gmail.com>:
>> > Hello,
>> >
>> > I as a part of my gsoc project(and on suggestion from my mentor )tried
>> > writing a bencmark for threading module.
>> > It uses "perf" to measure crearion of objects like
>> Thread,Semaphore,Locks
>> > etc. and few methods on them but It might lack on experience and the
>> > correctness of doing so.
>> >
>> > Here's the bm .
>> >
>> > And here are the results of a run on my local machine [specs]:
>> >
>> >
>> > python3 bm_threading.py
>> > .....................
>> > basic: Mean +- std dev: 14.4 us +- 0.4 us
>> > .....................
>> > condition: Mean +- std dev: 18.1 ns +- 1.1 ns
>> > .....................
>> > lock: Mean +- std dev: 18.1 ns +- 0.9 ns
>> > .....................
>> > rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
>> > .....................
>> > semaphore: Mean +- std dev: 298 ns +- 56 ns
>> > .....................
>> > timer: Mean +- std dev: 6.74 us +- 0.65 us
>> >
>> >
>> >
>> > python2 bm_threading.py
>> > .....................
>> > basic: Mean +- std dev: 20.5 us +- 1.1 us
>> > .....................
>> > condition: Mean +- std dev: 148 ns +- 2 ns
>> > .....................
>> > lock: Mean +- std dev: 21.1 ns +- 0.7 ns
>> > .....................
>> > rlock: Mean +- std dev: 151 ns +- 3 ns
>> > .....................
>> > semaphore: Mean +- std dev: 336 ns +- 4 ns
>> > .....................
>> > timer: Mean +- std dev: 12.5 us +- 0.7 us
>> >
>> > So suggestions?
>> >
>> > Thank You
>> >
>> > _______________________________________________
>> > Speed mailing list
>> > Speed(a)python.org
>> > https://mail.python.org/mailman/listinfo/speed
>> >
>>
>
>
[View Less]
Hello,
WHY-> Presently there are no benchmarks for concurrency,
METHOD USED-> This s
<https://gist.github.com/bhavishyagopesh/76529e7bf170948b4d3f3a51f6e6bc9a>cript
tries to benchmark "Concurrency" implemented using threading and
multiprocessing module.Actually "threads" in cpython are restricted by
"GIL",so it's not actually concurrent...On the other hand "multiprocessing"
module creates whole different processes but there is substaintial cost
involved in spawing a whole new …
[View More]process.
So the there is a trade-off involved which is evident as we increase
"CRUNCH_NO" variable.
ToNote-> Interestingly Here py2 looks faster.(actually by a considerable
margin)
I'm attaching graphs that compare py2 and py3.
IMPROVEMENTS-> More use cases(apart from number crunching)which could
verify this and reproduce similar thing.
SO Suggestions required?
thank you.
On Fri, Aug 11, 2017 at 4:09 AM, Victor Stinner <victor.stinner(a)gmail.com>
wrote:
> I don't understand what you are trying to test. For example, for a
> lock, it's very different if a single thread uses the lock, or if two
> threads use the lock. None of your benchmarks seem to measure
> concurrency.
>
> Victor
>
> 2017-08-11 0:33 GMT+02:00 Bhavishya <bhavishyagopesh(a)gmail.com>:
> > Hello,
> >
> > I as a part of my gsoc project(and on suggestion from my mentor )tried
> > writing a bencmark for threading module.
> > It uses "perf" to measure crearion of objects like
> Thread,Semaphore,Locks
> > etc. and few methods on them but It might lack on experience and the
> > correctness of doing so.
> >
> > Here's the bm .
> >
> > And here are the results of a run on my local machine [specs]:
> >
> >
> > python3 bm_threading.py
> > .....................
> > basic: Mean +- std dev: 14.4 us +- 0.4 us
> > .....................
> > condition: Mean +- std dev: 18.1 ns +- 1.1 ns
> > .....................
> > lock: Mean +- std dev: 18.1 ns +- 0.9 ns
> > .....................
> > rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
> > .....................
> > semaphore: Mean +- std dev: 298 ns +- 56 ns
> > .....................
> > timer: Mean +- std dev: 6.74 us +- 0.65 us
> >
> >
> >
> > python2 bm_threading.py
> > .....................
> > basic: Mean +- std dev: 20.5 us +- 1.1 us
> > .....................
> > condition: Mean +- std dev: 148 ns +- 2 ns
> > .....................
> > lock: Mean +- std dev: 21.1 ns +- 0.7 ns
> > .....................
> > rlock: Mean +- std dev: 151 ns +- 3 ns
> > .....................
> > semaphore: Mean +- std dev: 336 ns +- 4 ns
> > .....................
> > timer: Mean +- std dev: 12.5 us +- 0.7 us
> >
> > So suggestions?
> >
> > Thank You
> >
> > _______________________________________________
> > Speed mailing list
> > Speed(a)python.org
> > https://mail.python.org/mailman/listinfo/speed
> >
>
[View Less]
I don't understand what you are trying to test. For example, for a
lock, it's very different if a single thread uses the lock, or if two
threads use the lock. None of your benchmarks seem to measure
concurrency.
Victor
2017-08-11 0:33 GMT+02:00 Bhavishya <bhavishyagopesh(a)gmail.com>:
> Hello,
>
> I as a part of my gsoc project(and on suggestion from my mentor )tried
> writing a bencmark for threading module.
> It uses "perf" to measure crearion of objects like Thread,…
[View More]Semaphore,Locks
> etc. and few methods on them but It might lack on experience and the
> correctness of doing so.
>
> Here's the bm .
>
> And here are the results of a run on my local machine [specs]:
>
>
> python3 bm_threading.py
> .....................
> basic: Mean +- std dev: 14.4 us +- 0.4 us
> .....................
> condition: Mean +- std dev: 18.1 ns +- 1.1 ns
> .....................
> lock: Mean +- std dev: 18.1 ns +- 0.9 ns
> .....................
> rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
> .....................
> semaphore: Mean +- std dev: 298 ns +- 56 ns
> .....................
> timer: Mean +- std dev: 6.74 us +- 0.65 us
>
>
>
> python2 bm_threading.py
> .....................
> basic: Mean +- std dev: 20.5 us +- 1.1 us
> .....................
> condition: Mean +- std dev: 148 ns +- 2 ns
> .....................
> lock: Mean +- std dev: 21.1 ns +- 0.7 ns
> .....................
> rlock: Mean +- std dev: 151 ns +- 3 ns
> .....................
> semaphore: Mean +- std dev: 336 ns +- 4 ns
> .....................
> timer: Mean +- std dev: 12.5 us +- 0.7 us
>
> So suggestions?
>
> Thank You
>
> _______________________________________________
> Speed mailing list
> Speed(a)python.org
> https://mail.python.org/mailman/listinfo/speed
>
[View Less]
Hello,
I as a part of my gsoc project(and on suggestion from my mentor )tried
writing a bencmark for threading module.
It uses "perf" to measure crearion of objects like Thread,Semaphore,Locks
etc. and few methods on them but It might lack on experience and the
correctness of doing so.
Here's the bm
<https://gist.github.com/bhavishyagopesh/0790591474040575b5fa4470b9ea7b76> .
And here are the results of a run on my local machine [specs
<https://bhavishyagopesh.github.io/Second-Post/…
[View More]>]:
python3 bm_threading.py
.....................
basic: Mean +- std dev: 14.4 us +- 0.4 us
.....................
condition: Mean +- std dev: 18.1 ns +- 1.1 ns
.....................
lock: Mean +- std dev: 18.1 ns +- 0.9 ns
.....................
rlock: Mean +- std dev: 19.4 ns +- 1.6 ns
.....................
semaphore: Mean +- std dev: 298 ns +- 56 ns
.....................
timer: Mean +- std dev: 6.74 us +- 0.65 us
python2 bm_threading.py
.....................
basic: Mean +- std dev: 20.5 us +- 1.1 us
.....................
condition: Mean +- std dev: 148 ns +- 2 ns
.....................
lock: Mean +- std dev: 21.1 ns +- 0.7 ns
.....................
rlock: Mean +- std dev: 151 ns +- 3 ns
.....................
semaphore: Mean +- std dev: 336 ns +- 4 ns
.....................
timer: Mean +- std dev: 12.5 us +- 0.7 us
So suggestions?
Thank You
[View Less]