On Tue, May 05, 2015 at 05:10:42PM -0700, Bill McCloskey wrote: > On Tue, May 5, 2015 at 4:34 PM, Nicholas Nethercote <n.netherc...@gmail.com> > wrote: > > > "resident-unique" (only available on Linux, alas) is probably the most > > interesting measurement in this case. > > > > Usually I see resident-unique pretty consistently about 15-20MB higher than > explicit for content processes. > > I wonder if we track IPDL shared memory in about:memory. I would imagine we > do, but I don't see it anywhere.
As long as we exec() new processes for each new content process, we're going to have a lot of wasted memory because of relocations in data segments in libxul. Ironically, even though it doesn't have fork(), I think Windows doesn't have this problem (I /think/ it shares memory for relocated read-only data segments across processes, as long as it doesn't have to re-relocate because it couldn't get the same address.) Nuwa, aiui, can somewhat help here, but the possibly best option is actually to just not have a separate executable and fork() the main process (I didn't say this was going to be easy) Mike _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
