Another idea for you: > In our test, on the other hand, each thread queries its own set of > rows from the table. So, based on your explanation, here each thread > should end up with its own set of pages (more or less). However, even > in this case, I still observe a near sequential performance.
Even though each thread works basically with its own set of pages they all use the same cache and thus are subject to the combined limit on total number of pages in the cache. And to keep that combined limit working they need to maintain a combined LRU list of pages to know which pages should be evicted from cache first in case that's needed. Pavel On Mon, Aug 13, 2012 at 5:51 AM, Boris Kolpackov <bo...@codesynthesis.com> wrote: > Hi Richard, > > Richard Hipp <d...@sqlite.org> writes: > >> In shared-cache mode, the page cache is shared across threads. That means >> that each thread must acquire a mutex on the page cache in order to read >> it. Which means that access to the page cache is serialized. > > I just ran our concurrency test in different configurations and I > observer a similar behavior. That is, in the shared-cache mode, > read-only transactions on the same table are executed pretty much > sequentially. > > Also, your explanation doesn't feel quite satisfactory to me. In his > original email, Eric mentioned that his table contains just 50 rows. > Surely all this data would be loaded into the cache the first time > it is requested and then accessed concurrently by all the threads. > The only way I can see how the sequential performance could be > explained here is if the cache mutex did not distinguish between > readers and writers (which would seem to be a fairly natural thing > to do). > > In our test, on the other hand, each thread queries its own set of > rows from the table. So, based on your explanation, here each thread > should end up with its own set of pages (more or less). However, even > in this case, I still observe a near sequential performance. > > Any idea what else might be going on here? > > Boris > -- > Boris Kolpackov, Code Synthesis http://codesynthesis.com/~boris/blog > Compiler-based ORM system for C++ http://codesynthesis.com/products/odb > Open-source XML data binding for C++ http://codesynthesis.com/products/xsd > XML data binding for embedded systems http://codesynthesis.com/products/xsde > > _______________________________________________ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
