Python 3 vs. Java | Hacker News Reader

Python 3 vs. Java

benchmarksgame.alioth.debian.org

21 pointstvvocold12 comments

This is compared to the standard (CPython) implementation. PyPy is about 7 times faster, often uses less memory and generally optimizes pythonic code better, so would improve all three metrics. http://speed.pypy.org/

And ZipPy (Python on the JVM) handily beats PyPy on most benchmarks but obviously doesn't beat Java on the JVM for idiomatic code due to the better data structures available in Java Collections vs. Python's standard library.

I love Python, and was eager to read some great measurements for 3.4, but I am having trouble comprehending this chart. I read the chart:

Python 3 used what fraction? used how many times more?

Time-used |- |--- 25% median 75% ---| -| (Elapsed secs)

± ± 5× 27× 51× 77× 77×

to indicate that even the 25th percentile Python application required 5 times the amount of time to complete as its Java counterpart. It seems that only the pidigits implementation required less time to complete. How does that lead to the conclusion that Python is -approximately- faster?

Edit: fixed chart.

It's just poor wording. The "approximately" isn't "yes, it's approximately faster" but "here are the numbers -- approximately". Notice that it says the same thing when you reverse the Python vs Java comparison.

Hmmm.

2 : Approximately - Are the Python 3 programs faster?

I always wish these benchmarks had a "Conclusion" or "Discussion" section...

Those ones "have" : http://benchmarksgame.alioth.debian.org/dont-jump-to-conclus...

Particularly: "Last words -- A good starting point"

(Also see the homepage)

    tl;dr

    Measurement is highly specific -- the time taken for
    this benchmark task, by this toy program, with this
    programming language implementation, with these options,
    on this computer, with these workloads.

    Same toy program, same computer, same workload -- but
    much slower.

    Measurement is not prophesy.

I would be very interested to see performance with networking types of code. That would be of great interest to this audience.

So design the tasks, make the measurements, and publish your results.

There are a load of database related tasks I would love to benchmark, but I simply don't have the time. Posting a comment like that here, someone will often know of a similar benchmark and reply with it.

Reply on news.ycombinator.com