jdk-based logger tend to block
at java.util.logging.Logger.log(Logger.java:476) in concurrent
environments. This makes logging a main performance issue. To overcome
this problem, this is a add-on to jdk logging to put log entries on a
concurrent message queue and log the messages one by one using a
separate process.
- FTPClient uses the concurrent logging instead of the log4j logger
metadata and old rwi and for the citation index. The important
advancement is the separation of the citation index deletion because
that index is responsible for the linkdepth calculation. Now a search
index can be deleted without the citation index and that should cause
that less clickdepths must be post-processed.
This time it works like this:
- each peer provides its ranking information using the yacy/idx.json servlet
- peers with more than 1 GB ram will load this information from all other peers, combine that into one ranking table and store it locally. This happens during the start-up of the peer concurrently. The new generated file with the ranking information is at DATA/INDEX/<network>/QUEUES/hostIndex.blob
- this index is then computed to generate a new fresh ranking table. Peers which can calculate their own ranking table will do that every start-up to get latest feature updates until the feature is stable
- I computed new ranking tables as part of the distribition and commit it here also
- the YBR feature must be enabled manually by setting the YBR value in the ranking servlet to level 15. A default configuration for that is also in the commit but it does not affect your current installation only fresh peers
- a recursive block rank refinement is implemented but disabled at this point. it needs more testing
Please play around with the ranking settings and see if this helped to make search results better.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@7729 6c8d7289-2bf4-0310-a012-ef5d649a1542
when a search fails for a single url because the snippet cannot be generated, then the url reference is deleted from the index. This mechanism was redesign and enhanced. The process now also writes into the work tables into the table searchfl to prepare a re-indexing mechanism.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@7364 6c8d7289-2bf4-0310-a012-ef5d649a1542
- confirm that database objects that are not supposed to grow do not have a index memory management that is designed for growth
- changed index sorting method in such a way that it allocates less objects during quicksort
- database classes classes renaming (shorter, naming addresses that objects hold in RAM)
- added a large number of asserts to check if objects actually take the RAM that they should have
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@7019 6c8d7289-2bf4-0310-a012-ef5d649a1542
- added crawling queue sizes to /api/status_p.xml, syntax same as in queues_p.html
- fixed a bug in queue enumeration that caused a out of bounds exception
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6842 6c8d7289-2bf4-0310-a012-ef5d649a1542
- no direct deletion of referenced during search (shifted to time after search)
- bundling of all deletions for the references of a single word into one remove operation
- enhanced remove operation by caring that the collection is stored sorted (experimental)
- more String -> byte[] transition for search word lists
- clean up of unused code
- enhanced memory allocation of RowSet Objects (will use a little bit less memory which was wasted before)
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6823 6c8d7289-2bf4-0310-a012-ef5d649a1542
- enhanced the Balancer performance when building new domain stacks using a new Table buffer
- added the new Table buffer BufferedObjectIndex class
- changed order of access to LURL-read (prefereing segment over Crawl Queues) will reduced blocking time on balancer
- fixed PPM setting in Crawler_p servlet (had doubled values)
- reduced synchronization in IndexCell because it is not necessary: reduced blocking during indexing/merging/dumping
- removed did-you-mean cache in IndexCell because that caused too much overhead and more memory usage but was not very useful. This reduced also deadlocks that could be causes when searched are performed during indexing.
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6819 6c8d7289-2bf4-0310-a012-ef5d649a1542
The result should be a less usage of new String() and less memory usage (since a String-encapsulated byte[] has 40 bytes overhead)
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6815 6c8d7289-2bf4-0310-a012-ef5d649a1542
pass value as byte[], not as String. This should cause that less
byte[] <-> String conversions are made during time-critical tasks.
This redesign is not yet complete, more to come ..
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6775 6c8d7289-2bf4-0310-a012-ef5d649a1542
- reverted SVN 6766, this is too dangerous (may cause unexpected memory usage) and should not be necessary
git-svn-id: https://svn.berlios.de/svnroot/repos/yacy/trunk@6773 6c8d7289-2bf4-0310-a012-ef5d649a1542
orbiter