I agree that 16 transactions / second is much slower than what I usually see on linux, even with a slow file system configuration. But I still believe this is either disk or filesystem being slow. Could you please go through the file system benchmarking outlined on this wiki page: http://wiki.neo4j.org/content/Linux_Performance_Guide
Cheers, Tobias On Tue, Apr 19, 2011 at 10:11 PM, Bob Hutchison <hutch-li...@recursive.ca>wrote: > Hi Tobias, > > On 2011-04-19, at 1:48 AM, Tobias Ivarsson wrote: > > > Hi Bob, > > > > What happens here is that you perform a tiny operation in each > transaction, > > so what you are really testing here is how fast your file system can > flush, > > because with such tiny transactions all of the time is going to be spent > in > > transactional overhead (i.e. flushing transaction logs to the disk). > > > > The reason you see such large differences between Mac OS X and Linux is > > because Mac OS X cheats. Flushing a file (fdatasync) on Mac does pretty > much > > nothing. The only thing Mac OS X guarantees is that it will write the > data > > that you just flushed before it writes the next data block you flush, so > > called "ordered writes". This means that you could potentially get > data-loss > > on hard failure, but never in a way that makes your data internally > > inconsistent. > > Okay, that's makes some sense. Thanks for the information. > > > > > So to give a short answer to your questions: > > 1) The linux number is reasonable, Mac OS X cheats. > > 2) What you are testing is the write speed of your disk for writing small > > chunks of data. > > So you're thinking that 16 or 17 writes is what should be expected? > > Cheers, > Bob > > > > > Cheers, > > Tobias > > > > On Mon, Apr 18, 2011 at 10:57 PM, Bob Hutchison < > hutch-li...@recursive.ca>wrote: > > > >> Hi, > >> > >> Using Neo4j 1.3 and the Borneo (Clojure) wrapper I'm getting radically > >> different performance numbers with identical test code. > >> > >> The test is a simple-minded: create two nodes and a relation between > them. > >> No properties, no indexes, all nodes and relations are different. > >> > >> On OS X, it takes about 50s to perform that operation 50,000 times, < > 0.8s > >> to do it 500 times. It uses roughly 30-40% of one core to do this. > >> > >> On linux it takes about 30s to perform that operation 500 times. The CPU > >> usage is negligible (really negligible... almost none). > >> > >> I cannot explain the difference in behaviour. > >> > >> I have two questions: > >> > >> 1) is either of these a reasonable number? I hoping the OS X numbers are > >> not too fast. > >> > >> 2) any ideas as to what might be the cause of this? > >> > >> The Computers are comparable. The OS X is a 2.8 GHz i7, the linux box is > a > >> 3.something GHz Xeon (I don't remember the details). > >> > >> Thanks in advance for any help, > >> Bob > >> > >> _______________________________________________ > >> Neo4j mailing list > >> User@lists.neo4j.org > >> https://lists.neo4j.org/mailman/listinfo/user > >> > > > > > > > > -- > > Tobias Ivarsson <tobias.ivars...@neotechnology.com> > > Hacker, Neo Technology > > www.neotechnology.com > > Cellphone: +46 706 534857 > > _______________________________________________ > > Neo4j mailing list > > User@lists.neo4j.org > > https://lists.neo4j.org/mailman/listinfo/user > > ---- > Bob Hutchison > Recursive Design Inc. > http://www.recursive.ca/ > weblog: http://xampl.com/so > > > > > _______________________________________________ > Neo4j mailing list > User@lists.neo4j.org > https://lists.neo4j.org/mailman/listinfo/user > -- Tobias Ivarsson <tobias.ivars...@neotechnology.com> Hacker, Neo Technology www.neotechnology.com Cellphone: +46 706 534857 _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
