[ https://issues.apache.org/jira/browse/LUCENE-3892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13247175#comment-13247175 ]
Han Jiang edited comment on LUCENE-3892 at 4/5/12 12:05 PM: ------------------------------------------------------------ Hi Mike, I have changed my proposal a bit, but here are some questions here: {quote} * There are actually more than 2 codecs (eg we also have Lucene3x, SimpleText, sep/intblock (abstract), random codecs/postings formats for testing...), but our default codec now is Lucene40. {quote} Yes, but it seems that our baseline will be Lucene40 and Pulsing? Lucene3x is read-only, and other approaches are not productive. And, what is random codec? Does it mean to randomly pick up a codec for user? {quote} * I think you can use the existing abstract sep/intblock classes (ie, they implement layers like FieldsProducer/Consumer...), and then you can "just" implement the required methods (eg to encode/decode one int[] block). {quote} And this was my initial thought about the PForDelta interface: The class hierarchy will be as below (quite similar to pulsing): * PForDeltaPostingsFormat(extends PostingsFormat): It will define global behaviors such as file suffix, and provide customized FieldsWriter/Reader * PForDeltaFieldsWriter(extends FieldsConsumer): It will define how terms,docids,freq,offset are written into posting files. inner classes include: ** PForDeltaTermsConsumer(extends TermsConsumer) ** PForDeltaPostingsConsumer(extends PostingsConsumer) * PForDeltaFieldsReader(extends FieldsProducer): It will define how postings are read from index, and provide *Enum class to iterate docids, freqs etc. inner classes include: ** PForDeltaFieldsEnum(extends FieldsEnum) ** PForDeltaTermsEnum(extends TermsEnum) ** PForDeltaDocsEnum(extends DocsEnum) ** PForDeltaDocsAndPositonsEnum(extends DocsAndPostionsEnum) ** PForDeltaTerms(extends Terms) It seems that "BlockTermsReader/Writer" have already implement those subclasses, and we can just pass our Postings(Writer/Reader)Base as an argument, like PatchedFrameOfRefCodec::fieldsConsumer() does. Then, to introduce PForDeltaCodec into trunk, we should also introduce the "fixed codec"? Also, why isn't lucene40codec implemented with this line? {quote} * We may need to tune the skipper settings, based on profiling results from skip-intensive (Phrase, And) queries... since it's currently geared towards single-doc-at-once encoding. I don't think we should try to make a new skipper impl here... (there is a separate issue for that). {quote} I haven't investigated much about different kinds of queries. What are skipper settings? {quote} * Maybe explore the combination of pulsing and PForDelta codecs; seems like the combination of those two could be important, since for low docFreq terms, retrieving the docs is now more expensive... {quote} Yes, it seems that if PForDelta outperforms current approaches, a Pulsing version will work better. This feature will also come as "phase 2". was (Author: billy): {quote} * There are actually more than 2 codecs (eg we also have Lucene3x, SimpleText, sep/intblock (abstract), random codecs/postings formats for testing...), but our default codec now is Lucene40. {quote} Yes, but it seems that our baseline will be Lucene40 and Pulsing? Lucene3x is read-only, and other approaches are not productive. And, what is random codec? Does it mean to randomly pick up a codec for user? {quote} * I think you can use the existing abstract sep/intblock classes (ie, they implement layers like FieldsProducer/Consumer...), and then you can "just" implement the required methods (eg to encode/decode one int[] block). {quote} And this was my initial thought about the PForDelta interface: The class hierarchy will be as below (quite similar to pulsing): * PForDeltaPostingsFormat(extends PostingsFormat): It will define global behaviors such as file suffix, and provide customized FieldsWriter/Reader * PForDeltaFieldsWriter(extends FieldsConsumer): It will define how terms,docids,freq,offset are written into posting files. inner classes include: ** PForDeltaTermsConsumer(extends TermsConsumer) ** PForDeltaPostingsConsumer(extends PostingsConsumer) * PForDeltaFieldsReader(extends FieldsProducer): It will define how postings are read from index, and provide *Enum class to iterate docids, freqs etc. inner classes include: ** PForDeltaFieldsEnum(extends FieldsEnum) ** PForDeltaTermsEnum(extends TermsEnum) ** PForDeltaDocsEnum(extends DocsEnum) ** PForDeltaDocsAndPositonsEnum(extends DocsAndPostionsEnum) ** PForDeltaTerms(extends Terms) It seems that "BlockTermsReader/Writer" have already implement those subclasses, and we can just pass our Postings(Writer/Reader)Base as an argument, like PatchedFrameOfRefCodec::fieldsConsumer() does. Then, to introduce PForDeltaCodec into trunk, we should also introduce the "fixed codec"? Also, why isn't lucene40codec implemented with this line? {quote} * We may need to tune the skipper settings, based on profiling results from skip-intensive (Phrase, And) queries... since it's currently geared towards single-doc-at-once encoding. I don't think we should try to make a new skipper impl here... (there is a separate issue for that). {quote} I haven't investigated much about different kinds of queries. What are skipper settings? {quote} * Maybe explore the combination of pulsing and PForDelta codecs; seems like the combination of those two could be important, since for low docFreq terms, retrieving the docs is now more expensive... {quote} Yes, it seems that if PForDelta outperforms current approaches, a Pulsing version will work better? This feature will also come as "phase 2". > Add a useful intblock postings format (eg, FOR, PFOR, PFORDelta, > Simple9/16/64, etc.) > ------------------------------------------------------------------------------------- > > Key: LUCENE-3892 > URL: https://issues.apache.org/jira/browse/LUCENE-3892 > Project: Lucene - Java > Issue Type: Improvement > Reporter: Michael McCandless > Labels: gsoc2012, lucene-gsoc-12 > Fix For: 4.0 > > > On the flex branch we explored a number of possible intblock > encodings, but for whatever reason never brought them to completion. > There are still a number of issues opened with patches in different > states. > Initial results (based on prototype) were excellent (see > http://blog.mikemccandless.com/2010/08/lucene-performance-with-pfordelta-codec.html > ). > I think this would make a good GSoC project. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org