On Wed, Jul 29, 2009 at 5:55 AM, Robert Shield <robertshi...@chromium.org>wrote:
> Really nice doc! An attempt at potentially helpful comments follows: > > Regarding the class naming in the Out of process design, the convention > I've seen most often is to have FooHost classes run in the browser with Foo > classes in child processes. > > Proxy resolver process / crashes - the document mentions fallback if proxy > config is set to auto-detect. If the proxy config is instead set to use a > fixed pac url and the proxy resolver process repeatedly crashes or becomes > unresponsive could we offer a UI toast in the browser along the lines of > "Your network configuration is bad, would you like me to try a different > one?" upon which we would switch to auto-detect or direct connections? Or > would that be requiring too much of the user? > > IPC messages from proxy resolver process to the browser - What > functionality does Msg_Alert provide that Msg_OnError doesn't? > > Proxy resolver process / Alive - the proxy resolver process may decide to > that a PAC script is invalid while doing auto-detect. I interpreted the next > step to be notifying the browser process of the invalid script. Does it > signal to the browser that the script is invalid through Msg_OnError or > Msg_Alert or through some special invocation of > Msg_GetProxyForURL_Complete? > > Proxy resolver process / Unresponsive - the section just below on Request > timeouts seems like it answers the TODO(eroman): How do we measure > "unresponsive"? question. Request timeouts on the browser process -> proxy > resolver process requests seem pretty important to have for v1. > > Performance - additional latency is (1 + <number of DNS resolves>) * > kIpcRoundtripLatency. Totally premature question tainted by a dash of > sandboxing ignorance: is it possible to embed the proxy resolver process in > a sandbox that _does_ allow network requests (but is otherwise similar to > the renderer sandbox)? This would chop down the latency although it would > require maintaining a separate dns resolver cache from the browser. > Isn't IPC latency on the order of microseconds? (Or less?) If so, I'd imagine not sharing the dns resolver cache would be a net loss. It is an interesting question in terms of how easily it'd be to stick the entire network stack in such a helper process, though. J > On Wed, Jul 29, 2009 at 6:29 AM, Eric Roman <ero...@chromium.org> wrote: > >> >> Here is a design document for http://crbug.com/11746 >> >> >> http://sites.google.com/a/chromium.org/dev/developers/design-documents/out-of-process-v8-pac >> >> Feedback welcome. >> >> >> > > > > --~--~---------~--~----~------------~-------~--~----~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~----------~----~----~----~------~----~------~--~---
