Hi Daniel: What you wrote seems reasonable. I ran the test again for a much longer period in an i86 environment and no memory allocation failures occurred. Of course, the uClinux environment has much less spare memory-- a little more than 2 MB. I think it is relevant that no other tasks--other than uClinux system background tasks, were running when I ran the test on uClinux.
Best Regards, Paul R. Daniel Stenberg wrote: > On Thu, 9 Sep 2010, Paul Romero wrote: > > > Thanks. In fact the test does free all connections in between transactions. > > libssh2 is not easy on the heap. It allocates and frees a lot of memory chunks > and if your app does it as well, chances are that the memory pools are so > fragmented that there isn't any large enough blocks left when libssh2 wants > one. > > I've been wanting to reduce the amount of mallocs but I've not yet gotten > around actually making that reality... > > -- > > / daniel.haxx.se > _______________________________________________ > libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel -- Paul Romero RCOM Communications Software Phone/Fax: (510)339-2628 E-Mail: [email protected] _______________________________________________ libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
