Le 3/17/13 11:56 PM, Emmanuel Lécharny a écrit : > Hi, > > I just completed the code that write data on a Managed BTree (ie a BTree > which uses SoftReference to store data, instead of holding them in > memory), and here are the very preliminary results (no optimization > whatsoever). These tests are ran on my 3 years old MacbookPro, using Java 6. > > I've injected 100 000 <Long, String> elements, into a BTree which > accepts 256 elements per page, and with a RM using 4096 bytes size pages. > > - the resulting file is roughly 1.2 Gb (I hav'nt implemented yet the > reuse of not used pages) > - I can write around 3400 tuple per second > - I can check the existence of keys at a 8200 keys per second > - Once the first check is done, I'm able to check the existence of > 10900 keys per second
Those numbers are wrong (at least for the exist() call ), as I was doing the operation twice (on call to exists() and one call to get() in the loop), and there were a bug in the way we stored the elements in SoftReferences : we were always reading them from disk. - in case the elements are not in memory, then we can read around 18 000 elements per second - once teh SoftReferences are loaded, we can read up to 588 000 elements per second. -- Regards, Cordialement, Emmanuel Lécharny www.iktek.com --------------------------------------------------------------------- To unsubscribe, e-mail: labs-unsubscr...@labs.apache.org For additional commands, e-mail: labs-h...@labs.apache.org
