[
https://issues.apache.org/jira/browse/LUCENE-1990?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Toke Eskildsen updated LUCENE-1990:
-----------------------------------
Attachment: LUCENE-1990-te20100226.patch
Now we're getting somewhere. I finally squashed the persistence bug and the
tests has been turned up another notch. Everything seems to run as it should.
Pending issues, as I see them:
- Review of the code
- Should we make a MutableWriter?
- Should we drop support for aligned?
The last one is interesting. The code for getting a value from aligned uses
devision and a single RAM-request:
{code}
public long get(final int index) {
final int blockPos = index / valuesPerBlock;
final int bitPos = (index - (blockPos * valuesPerBlock)) * bitsPerValue;
return (blocks[blockPos] >>> shifts[bitPos]) & readMask;
{code}
where the code for packed uses shift and two RAM-requests:
{code}
final long majorBitPos = index * bitsPerValue;
final int elementPos = (int)(majorBitPos >>> BLOCK_BITS); // / BLOCK_SIZE
final int bitPos = (int)(majorBitPos & MOD_MASK); // % BLOCK_SIZE);
final int base = bitPos * FAC_BITPOS;
return ((blocks[elementPos] << shifts[base]) >>> shifts[base+1]) |
((blocks[elementPos+1] >>> shifts[base+2]) & readMasks[bitPos]);
{code}
I have done some tests (see the TODO-file in the attached patch) and on 64 bit
machines, the difference in access-speed for aligned vs. packed is not that
great and not always in favor of aligned. Probably because some space is wasted
and the RAM-cache is not so well utilized. If this is also the case for 32 bit
machines, I vote for removing aligned and only used packed with the
special-case optimizations direct8, direct16, direct32 and direct64. This would
also mean that there is only one persistent format.
{code}
java -cp lucene-core-3.1-dev.jar
org.apache.lucene.util.packed.PackedIntsPerformance
{code}
Runs throught the performance tests and delivers a simple report, so it should
be very easy to test on different platforms. It only measures access speed.
I consider this patch ready for review and concentrate on other matters until I
hear more.
> Add unsigned packed int impls in oal.util
> -----------------------------------------
>
> Key: LUCENE-1990
> URL: https://issues.apache.org/jira/browse/LUCENE-1990
> Project: Lucene - Java
> Issue Type: Improvement
> Components: Index
> Reporter: Michael McCandless
> Priority: Minor
> Attachments: LUCENE-1990-te20100122.patch,
> LUCENE-1990-te20100210.patch, LUCENE-1990-te20100212.patch,
> LUCENE-1990-te20100223.patch, LUCENE-1990-te20100226.patch,
> LUCENE-1990.patch, LUCENE-1990_PerformanceMeasurements20100104.zip
>
>
> There are various places in Lucene that could take advantage of an
> efficient packed unsigned int/long impl. EG the terms dict index in
> the standard codec in LUCENE-1458 could subsantially reduce it's RAM
> usage. FieldCache.StringIndex could as well. And I think "load into
> RAM" codecs like the one in TestExternalCodecs could use this too.
> I'm picturing something very basic like:
> {code}
> interface PackedUnsignedLongs {
> long get(long index);
> void set(long index, long value);
> }
> {code}
> Plus maybe an iterator for getting and maybe also for setting. If it
> helps, most of the usages of this inside Lucene will be "write once"
> so eg the set could make that an assumption/requirement.
> And a factory somewhere:
> {code}
> PackedUnsignedLongs create(int count, long maxValue);
> {code}
> I think we should simply autogen the code (we can start from the
> autogen code in LUCENE-1410), or, if there is an good existing impl
> that has a compatible license that'd be great.
> I don't have time near-term to do this... so if anyone has the itch,
> please jump!
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
