Not to mention that hotspot takes time to warm up. You should run it
twice within the same JVM, once for warmup, and once to measure results.
Doug Cutting wrote:
> Andi,
>
> I think these numbers could still easily be dominated by class loading.
> Java classes are loaded on demand, and there are many classes which are
> probably loaded after the IndexFiles demo records the start time. I'd
> guess that there is no class loading with gcj. It would be better to
> use a benchmark that takes at least 10-20 seconds. For example, try
> using IndexFiles or IndexHTML on the JDK documentation.
>
> Cheers,
>
> Doug
>
> Andi Vajda wrote:
>
>> The numbers are the ones that are displayed by the lucene
>> org.apache.lucene.demo.IndexFiles demo, I assume the clock is started
>> after
>> the VM has loaded.
>>
>> Andi..
>>
>> On Wed, 3 Dec 2003, Dmitry Serebrennikov wrote:
>>
>>
>>> Andi Vajda wrote:
>>>
>>>
>>>> I got the latest lucene to compile and run its demo on Linux (redhat
>>>> 9) and
>>>> Mac OS X Panther 10.3.1 using gcc 3.3.2's gcj, compiling to native
>>>> executables.
>>>>
>>>> The demo seems to run. I didn't run the unittests since I didn't
>>>> attempt to
>>>> compile junit with gcj yet.
>>>>
>>>> Performance seems way faster than the java VM version but somewhat
>>>> slower than
>>>> clucene's C++ port of lucene, indexing a small set of files. I say
>>>> somewhat as
>>>> my set is not that large and my measurements not very thorough, but
>>>> here are
>>>> some numbers for indexing 116k of text in 14 files, as measured on a
>>>> Powerbook
>>>> 1Ghz/1GB running Mac OS X Panther 10.3.1:
>>>>
>>>> lucene 1.3 RC3 with java 1.4.1: 1169 ms
>>>> lucene 1.3 RC3 with gcj : 467 ms
>>>> clucene 0.8.9 : 365 ms
>>>>
>>>
>>> This is very interesting! I didn't even know about the clucene project,
>>> but now I'll definetely have to take a look! I was beginning to think
>>> about doing this myself, but if the code already exists - so much the
>>> better. gcj sounds great too.
>>> Are these times for pure indexing, or is this a time for the program to
>>> run. If so, the extra 1 second could just be due to the VM startup
>>> time... The numbers would be much less impressive if that was the case.
>>> Also, if you have numbers for a larger dataset, please post them here. I
>>> am very interested in these results, and I think many others might be
>>> also.
>>>
>>> Thanks.
>>> Dmitry.
>>>
>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: lucene-dev-unsubscribe@jakarta.apache.org
>>> For additional commands, e-mail: lucene-dev-help@jakarta.apache.org
>>>
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: lucene-dev-unsubscribe@jakarta.apache.org
>> For additional commands, e-mail: lucene-dev-help@jakarta.apache.org
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: lucene-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: lucene-dev-help@jakarta.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: lucene-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: lucene-dev-help@jakarta.apache.org
|