yacy_search_server/source/de/anomic/search/ReferenceOrder.java
orbiter 0d363a94d7 more performance hacks
this makes YaCy search results VERY fast for all verify=false search cases
and it enhances the search speed also for all other snippet-fetch cases.
With this change my peer performed 100 Queries Per Second (!!!) while doing 10 queries simultanously (!!!)
in an intranet index of 20000 URLs on my 16-core Mac

Check this yourself by doing:
cd bin
./searchtestmulti.sh
after finishing the run, divide 1000 by the given time per query (which is the qps for one thread)
and then multiply again by 10 (because 10 search threads has been started)

git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@7231 6c8d7289-2bf4-0310-a012-ef5d649a1542
2010-10-09 08:55:57 +00:00

248 lines
13 KiB
Java

// ReferenceOrder.java
// (C) 2007 by Michael Peter Christen; mc@yacy.net, Frankfurt a. M., Germany
// first published 07.11.2007 on http://yacy.net
//
// This is a part of YaCy, a peer-to-peer based web search engine
//
// $LastChangedDate$
// $LastChangedRevision$
// $LastChangedBy$
//
// LICENSE
//
// This program is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
package de.anomic.search;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.concurrent.BlockingQueue;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.Semaphore;
import net.yacy.document.Condenser;
import net.yacy.document.LargeNumberCache;
import net.yacy.kelondro.data.meta.DigestURI;
import net.yacy.kelondro.data.word.WordReference;
import net.yacy.kelondro.data.word.WordReferenceRow;
import net.yacy.kelondro.data.word.WordReferenceVars;
import net.yacy.kelondro.logging.Log;
import net.yacy.kelondro.order.Bitfield;
import net.yacy.kelondro.rwi.ReferenceContainer;
import net.yacy.kelondro.util.ScoreCluster;
public class ReferenceOrder {
private static int cores = Runtime.getRuntime().availableProcessors();
private int maxdomcount;
private WordReferenceVars min, max;
private final ScoreCluster<String> doms; // collected for "authority" heuristic
private final RankingProfile ranking;
private final String language;
public ReferenceOrder(final RankingProfile profile, String language) {
this.min = null;
this.max = null;
this.ranking = profile;
this.doms = new ScoreCluster<String>();
this.maxdomcount = 0;
this.language = language;
}
public BlockingQueue<WordReferenceVars> normalizeWith(final ReferenceContainer<WordReference> container) {
LinkedBlockingQueue<WordReferenceVars> out = new LinkedBlockingQueue<WordReferenceVars>();
int threads = cores + 1;
if (container.size() < 20) threads = 2;
Thread distributor = new NormalizeDistributor(container, out, threads);
distributor.start();
// return the resulting queue while the processing queues are still working
return out;
}
private final class NormalizeDistributor extends Thread {
ReferenceContainer<WordReference> container;
LinkedBlockingQueue<WordReferenceVars> out;
private int threads;
public NormalizeDistributor(ReferenceContainer<WordReference> container, LinkedBlockingQueue<WordReferenceVars> out, int threads) {
this.container = container;
this.out = out;
this.threads = threads;
}
@Override
public void run() {
// transform the reference container into a stream of parsed entries
BlockingQueue<WordReferenceVars> vars = WordReferenceVars.transform(container);
// start the transformation threads
Semaphore termination = new Semaphore(this.threads);
NormalizeWorker[] worker = new NormalizeWorker[this.threads];
for (int i = 0; i < this.threads; i++) {
worker[i] = new NormalizeWorker(out, termination);
worker[i].start();
}
// fill the queue
WordReferenceVars iEntry;
int p = 0;
try {
while ((iEntry = vars.take()) != WordReferenceVars.poison) {
worker[p % this.threads].add(iEntry);
p++;
}
} catch (InterruptedException e) {
}
// insert poison to stop the queues
for (int i = 0; i < this.threads; i++) worker[i].add(WordReferenceVars.poison);
}
}
/**
* normalize ranking: find minimum and maximum of separate ranking criteria
*/
private class NormalizeWorker extends Thread {
private final BlockingQueue<WordReferenceVars> out;
private final Semaphore termination;
private final BlockingQueue<WordReferenceVars> decodedEntries;
public NormalizeWorker(final BlockingQueue<WordReferenceVars> out, Semaphore termination) {
this.out = out;
this.termination = termination;
this.decodedEntries = new LinkedBlockingQueue<WordReferenceVars>();
}
public void add(WordReferenceVars entry) {
try {
decodedEntries.put(entry);
} catch (InterruptedException e) {
}
}
public void run() {
try {
WordReferenceVars iEntry;
Map<String, Integer> doms0 = new HashMap<String, Integer>();
String dom;
Integer count;
final Integer int1 = 1;
while ((iEntry = decodedEntries.take()) != WordReferenceVars.poison) {
// find min/max
if (min == null) min = iEntry.clone(); else min.min(iEntry);
if (max == null) max = iEntry.clone(); else max.max(iEntry);
out.put(iEntry); // must be after the min/max check to prevent that min/max is null in cardinal()
// update domcount
dom = new String(iEntry.metadataHash(), 6, 6);
count = doms0.get(dom);
if (count == null) {
doms0.put(dom, int1);
} else {
doms0.put(dom, LargeNumberCache.valueOf(count.intValue() + 1));
}
}
// update domain score
Map.Entry<String, Integer> entry;
final Iterator<Map.Entry<String, Integer>> di = doms0.entrySet().iterator();
while (di.hasNext()) {
entry = di.next();
doms.addScore(entry.getKey(), (entry.getValue()).intValue());
}
if (!doms.isEmpty()) maxdomcount = doms.getMaxScore();
} catch (InterruptedException e) {
Log.logException(e);
} catch (Exception e) {
Log.logException(e);
} finally {
// insert poison to signal the termination to next queue
try {
this.termination.acquire();
if (this.termination.availablePermits() == 0) this.out.put(WordReferenceVars.poison);
} catch (InterruptedException e) {}
}
}
}
public int authority(final byte[] urlHash) {
return (doms.getScore(new String(urlHash, 6, 6)) << 8) / (1 + this.maxdomcount);
}
/**
* return the ranking of a given word entry
* @param t
* @return a ranking: the higher the number, the better is the ranking
*/
public long cardinal(final WordReferenceVars t) {
//return Long.MAX_VALUE - preRanking(ranking, iEntry, this.entryMin, this.entryMax, this.searchWords);
// the normalizedEntry must be a normalized indexEntry
final Bitfield flags = t.flags();
assert min != null;
assert max != null;
assert t != null;
assert ranking != null;
final long tf = ((max.termFrequency() == min.termFrequency()) ? 0 : (((int)(((t.termFrequency()-min.termFrequency())*256.0)/(max.termFrequency() - min.termFrequency())))) << ranking.coeff_termfrequency);
//System.out.println("tf(" + t.urlHash + ") = " + Math.floor(1000 * t.termFrequency()) + ", min = " + Math.floor(1000 * min.termFrequency()) + ", max = " + Math.floor(1000 * max.termFrequency()) + ", tf-normed = " + tf);
int maxmaxpos = max.maxposition();
int minminpos = min.minposition();
final long r =
((256 - DigestURI.domLengthNormalized(t.metadataHash())) << ranking.coeff_domlength)
+ ((ranking.coeff_ybr > 12) ? ((256 - (RankingProcess.ybr(t.metadataHash()) << 4)) << ranking.coeff_ybr) : 0)
+ ((max.urlcomps() == min.urlcomps() ) ? 0 : (256 - (((t.urlcomps() - min.urlcomps() ) << 8) / (max.urlcomps() - min.urlcomps()) )) << ranking.coeff_urlcomps)
+ ((max.urllength() == min.urllength() ) ? 0 : (256 - (((t.urllength() - min.urllength() ) << 8) / (max.urllength() - min.urllength()) )) << ranking.coeff_urllength)
+ ((maxmaxpos == minminpos ) ? 0 : (256 - (((t.minposition() - minminpos ) << 8) / (maxmaxpos - minminpos) )) << ranking.coeff_posintext)
+ ((max.posofphrase() == min.posofphrase()) ? 0 : (256 - (((t.posofphrase() - min.posofphrase() ) << 8) / (max.posofphrase() - min.posofphrase()) )) << ranking.coeff_posofphrase)
+ ((max.posinphrase() == min.posinphrase()) ? 0 : (256 - (((t.posinphrase() - min.posinphrase() ) << 8) / (max.posinphrase() - min.posinphrase()) )) << ranking.coeff_posinphrase)
+ ((max.distance() == min.distance() ) ? 0 : (256 - (((t.distance() - min.distance() ) << 8) / (max.distance() - min.distance()) )) << ranking.coeff_worddistance)
+ ((max.virtualAge() == min.virtualAge()) ? 0 : (((t.virtualAge() - min.virtualAge() ) << 8) / (max.virtualAge() - min.virtualAge()) ) << ranking.coeff_date)
+ ((max.wordsintitle() == min.wordsintitle()) ? 0 : (((t.wordsintitle() - min.wordsintitle() ) << 8) / (max.wordsintitle() - min.wordsintitle()) ) << ranking.coeff_wordsintitle)
+ ((max.wordsintext() == min.wordsintext()) ? 0 : (((t.wordsintext() - min.wordsintext() ) << 8) / (max.wordsintext() - min.wordsintext()) ) << ranking.coeff_wordsintext)
+ ((max.phrasesintext() == min.phrasesintext()) ? 0 : (((t.phrasesintext()- min.phrasesintext() ) << 8) / (max.phrasesintext()- min.phrasesintext()) ) << ranking.coeff_phrasesintext)
+ ((max.llocal() == min.llocal()) ? 0 : (((t.llocal() - min.llocal() ) << 8) / (max.llocal() - min.llocal()) ) << ranking.coeff_llocal)
+ ((max.lother() == min.lother()) ? 0 : (((t.lother() - min.lother() ) << 8) / (max.lother() - min.lother()) ) << ranking.coeff_lother)
+ ((max.hitcount() == min.hitcount()) ? 0 : (((t.hitcount() - min.hitcount() ) << 8) / (max.hitcount() - min.hitcount()) ) << ranking.coeff_hitcount)
+ tf
+ ((ranking.coeff_authority > 12) ? (authority(t.metadataHash()) << ranking.coeff_authority) : 0)
+ ((flags.get(WordReferenceRow.flag_app_dc_identifier)) ? 255 << ranking.coeff_appurl : 0)
+ ((flags.get(WordReferenceRow.flag_app_dc_title)) ? 255 << ranking.coeff_app_dc_title : 0)
+ ((flags.get(WordReferenceRow.flag_app_dc_creator)) ? 255 << ranking.coeff_app_dc_creator : 0)
+ ((flags.get(WordReferenceRow.flag_app_dc_subject)) ? 255 << ranking.coeff_app_dc_subject : 0)
+ ((flags.get(WordReferenceRow.flag_app_dc_description)) ? 255 << ranking.coeff_app_dc_description : 0)
+ ((flags.get(WordReferenceRow.flag_app_emphasized)) ? 255 << ranking.coeff_appemph : 0)
+ ((flags.get(Condenser.flag_cat_indexof)) ? 255 << ranking.coeff_catindexof : 0)
+ ((flags.get(Condenser.flag_cat_hasimage)) ? 255 << ranking.coeff_cathasimage : 0)
+ ((flags.get(Condenser.flag_cat_hasaudio)) ? 255 << ranking.coeff_cathasaudio : 0)
+ ((flags.get(Condenser.flag_cat_hasvideo)) ? 255 << ranking.coeff_cathasvideo : 0)
+ ((flags.get(Condenser.flag_cat_hasapp)) ? 255 << ranking.coeff_cathasapp : 0)
+ ((patchUK(t.language).equals(this.language)) ? 255 << ranking.coeff_language : 0)
+ ((DigestURI.probablyRootURL(t.metadataHash())) ? 15 << ranking.coeff_urllength : 0);
//if (searchWords != null) r += (yacyURL.probablyWordURL(t.urlHash(), searchWords) != null) ? 256 << ranking.coeff_appurl : 0;
return r; // the higher the number the better the ranking.
}
private static final String patchUK(String l) {
// this is to patch a bad language name setting that was used in 0.60 and before
if (l == null || l.equals("uk")) return "en"; else return l;
}
}