- each cache can now allocate as much memory as is available
- no more fixed limits
- replaced old performance memory monitor by new one
- added supervision methods as static functions into the classes that provide cache functionality
- steering of ram allocation is done with two simple limits that are ram availability-relative
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@3434 6c8d7289-2bf4-0310-a012-ef5d649a1542
- new domains now get a chance to get crawled early
- less IO operations
- new balancing method
- better dump order at shutdown time
- bugfixes regarding not found url hashes (no more superfluous cache kill)
- domain access time is now shared over all balancer stacks
- viewing the stack does no more disturbish the balancing algorithm that much
- intelligent selection of best next domain using domain access times
- extra double-check (to double-check the double-check)
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@3384 6c8d7289-2bf4-0310-a012-ef5d649a1542
redesign for better IO performance
enhanced database seek-time by avoiding write operations at distant
positions of a database file. until now, a USEDC counter was written
at the head-section of a kelondroRecords database file (which is the
basic data structure of all kelondro database files) to store the
actual number of records that are contained in the database. Now, this
value is computed from the database file size. This is either done
only once at start-time, or continuously when run in asserts enabled.
The counter is then updated only in RAM, and written at close of the
file. If the close fails, the correct number can be computed from the
file size, and if this is not equal to the stored number it is a strong
evidence that YaCY was not shut down properly.
To preserve consistency, the complete storage-routine had to be re-written.
Another change enhances read of nodes in some cases, where the data-tail
can be read together with the data-head. This saves another IO lookup during
each DB node fetch.
Includes also many small bugfixes.
IF ANYTHING GOES WRONG, ALL YOUR DATA IS LOST: PLEASE MAKE A BACK-UP
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@3375 6c8d7289-2bf4-0310-a012-ef5d649a1542
- controlled object order for all database tables
- migrated DHT position computation to correct base64-decoded values
this also closed the 'gaps' in the dht positions
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@3049 6c8d7289-2bf4-0310-a012-ef5d649a1542
* replaced kelondroTree for NURLs by kelondroFlex
* replaced kelondroTree for EURLs by kelondroFlex
take care, may be very buggy
please finish crawls before updating. crawls will be lost.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@2452 6c8d7289-2bf4-0310-a012-ef5d649a1542
* cleaned up code
* added unit test code
* migrated ranking RCI computation to kelondroFlex and kelondroCollectionIndex
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@2414 6c8d7289-2bf4-0310-a012-ef5d649a1542
* adopted all code to use the declaration form of kelondroRow
* fixed a bug in kelondroRow which caused wrong parsing of encoding type
* the bug caused bad database behaviour in new indexCollection data structure.
because of this bug, all test databases are now already void. A new database is created
* the kelondroFlexTable and indexCollection data structures now store a declaration of the row definition
into a properties file along the database files.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@2375 6c8d7289-2bf4-0310-a012-ef5d649a1542
this was done because testing showed that cache-delete operations
slowed down record access most, even more that actual IO operations.
Cache-delete operations appeared when entries were shifted from low-priority
positions to high-priority positions. During a fill of x entries to a database,
x/2 delete situation happen which caused two or more delete operations.
removing the cache control means that these delete operations are not
necessary any more, but it is more difficult to decide which cache elements
shall be removed in case that the cache is full. There is not yet a stable
solution for this case, but the advantage of a faster cache is more important
that the flush problem.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@2244 6c8d7289-2bf4-0310-a012-ef5d649a1542
various checks like the blacklist check or the robots.txt disallow check are now
done by a separate thread to unburden the indexer thread(s)
TODO: maybe we have to introduce a threadpool here if it turn out that this single
thread is a bottleneck because of the time consuming robots.txt downloads
*) improved index transfer
The index selection and transmission is done in parallel now to improve index
transfer performance.
TODO: maybe we could speed up performance by unsing multiple transmission threads in
parallel instead of only a single one.
*) gzip encoded post requests
it is now configureable if a gzip encoded post request should be send on
intex transfer/distribution
*) storage Peer (very experimentell and not optimized yet)
Now it's possible to send the result of the yacy indexer thread to a remote peer
istead of storing the indexed words locally.
This could be done by setting the property "storagePeerHash" in the yacy config file
- Please note that if the index transfer fails, the index ist stored locally.
- TODO: currently this index transfer is done by the indexer thread.
To seedup the indexer
a) this transmission should be done in parallel and
b) multiple chunks should be bundled and transfered together
*) general performance improvements
- better memory cleanup after http request processing has finished
- replacing some string concatenations with stringBuffers
- replacing BufferedInputStreams with serverByteBuffer
- replacing vectors with arraylists wherever possible
- replacing hashtables with hashmaps wherever possible
This was done because function calls to verctor or hashtable functions
take 3 time longer than calls to functions of arraylists or hashmaps.
TODO: we should take a look on the class serverObject which is inherited from hashmap
Do we realy need a synchronization for this class?
TODO: replace arraylists with linkedLists if random access to the list elements is not needed
*) Robots Parser supports if-modified-since downloads now
If the downloaded robots.txt file is older than 7 days the robots parser tries to
download the robots.txt with the if-modified-since header to avoid unnecessary downloads
if the file was not changed. Additionally the ETag header is used to detect changes.
*) Crawler: better handling of unsupported mimeTypes + FileExtension
*) Bugfix: plasmaWordIndexEntity was not closed correctly in
- query.java
- plasmaswitchboard.java
*) function minimizeUrlDB added to yacy.java
this function tests the current urlHashDB for unused urls
ATTENTION: please don't use this function at the moment because
it causes the wordIndexDB to flush all words into the
word directory!
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@853 6c8d7289-2bf4-0310-a012-ef5d649a1542
Crawler StartURLs will now also added to the errorURL-DB if an error occures on this url
*) kelondroStack.java, plasmaSwitchboardQueue.java
Adding method which returns a list of all entries in the queue. This list is used by IndexCreate_p.java
instead of an iterator to display the indexing-list.
Advantages: avoid concurrent modifications of the list while displaying it.
Speedup because now we have to access only one sync function instead of multiple ones
(one for each entry)
*) IndexCreateIndexingQueue_p.java
Using new list() function of plasmaSwitchboardQueue
*) httpdFileHandler.java
If a servelet returns the special value "LOCATION" the httpFileHandler does a Redirection of
the Browser to the URL specified by the servelet. This can e.g. be used when a http get request is
used insead of a post request, but a refresh should not be allowed.
*) IndexCreateWWWLocalQueue_p.html
Now it's possible to delete single entries of the local crawler queue
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@626 6c8d7289-2bf4-0310-a012-ef5d649a1542
orbiter