On Tue, May 31, 2016 at 05:01:34PM -0700, L. David Baron wrote: > On Tuesday 2016-05-31 12:50 -0400, Benjamin Smedberg wrote: > > It's likely that this particular report is running out of VM, yes. jemalloc > > allocates new memory chunks in large blocks (1MB?), and with only 122MB of > > VM it's likely that a lot of that is inaccessible, either because of > > fragmentation or because sites are allocating VM blocks of less than 64k, > > which is the allocation resolution of Windows VM. > > > > If you look at the raw dump tab, you'll see: > > > > "largest_free_vm_block": "0xf0000" - which is 983040 bytes, less than 1MB. > > Would it make sense for jemalloc to try allocating memory in smaller > chunks when large ones aren't available?
The way jemalloc (currently) works doesn't allow that. Making chunks smaller is one option. Future versions might even remove the notion of chunks (https://github.com/jemalloc/jemalloc/issues/360) Mike _______________________________________________ dev-platform mailing list [email protected] https://lists.mozilla.org/listinfo/dev-platform
