Hmm, very interesting. So in a nutshell : the use of a memory mapped tmpfs for /run/shm regions is making the memory problem worse. Maybe those could go to an app specific directory, and tax the local filesystem instead. Modulo the performance impact.
Chris: what do you think? On Tue, Sep 22, 2015 at 7:02 PM, Jean-Baptiste Lallement < [email protected]> wrote: > The way chromium uses shm is a problem on devices with limited resources > (small amount of physical memory and small swap due to disk space > limitation) To illustrate the problem I attached a profile of the memory > usage while browsing web sites. > > For this test I browsed mobile.lemonde.fr for approximately 5 minutes, > staying on the same website and using links to various news articles, > then I switched to fr.euronews.com (open from the News) scope. The way > this site is designed and rendered uses lot of resources. I navigated on > euronews for approx. 5 minutes too using links to articles on the right > of the page. > > The graph at the top shows the memory consumption when chromium stores its > files in /run/shm (of type tmpfs). > The graph at the bottom shows the memory consumption when chromium stores > its files on disk (I bindmounted a directory on disk on top of /run/shm) > > On the first graph you see that when you open Euronews ( around 300s) > there is immediately lot of swap allocated, then after 1 minute OOM is > triggered and kills the renderer process, after 2 minutes of browsing it > kills it again (visible on the graph when mem_used suddenly drops) > > On the second graph, the profile is very similar when you browse > lemonde.fr,. When I opened Euronews (360s) only a little amount of swap > is used , I continued navigating on Euronews, OOM never went off and > memory consumption stayed at a reasonable level (The graph drops at the > end when I closed the browser) > > Although, with shm on disk, there is an impact on the performance of the > browser and rendering sometimes become slow. > > ** Attachment added: "phone_memusage.png" > > https://bugs.launchpad.net/ubuntu/+source/oxide-qt/+bug/1497924/+attachment/4471487/+files/phone_memusage.png > > -- > You received this bug notification because you are a bug assignee. > Matching subscriptions: oxide in Ubuntu > https://bugs.launchpad.net/bugs/1497924 > > Title: > oxide-renderer heavily uses /run/shm > > To manage notifications about this bug go to: > > https://bugs.launchpad.net/canonical-devices-system-image/+bug/1497924/+subscriptions > -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1497924 Title: oxide-renderer heavily uses /run/shm To manage notifications about this bug go to: https://bugs.launchpad.net/canonical-devices-system-image/+bug/1497924/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
