indeed without nagle disabled (which is what nodelay does) a pipeline
can be interrupted for a full rtt waiting for an ack, which defeats
the purpose of pipelining. Other parts of FF can just coalesce the
writes internally, but sometimes that is not possible for pipelines
which may not have an opportunity for a pipelined request when the
"runt" packet goes out, but discovers such an opportunity a long time
before the rtt is expired. In other places (such as long posts) we do
a fine job of coalescing already, so this should not result in any
extra packets on the wire, just improved latency.
--HG--
extra : rebase_source : e333e36fe31b63d17c287ac6e876f27fdb69ce5b
raise sockettransportservice max sockets to 550 from 50 for linux, os
x, and windows >= xp. This does not change the default http
max-connections config (which remains at 30), but does allow
configurations above 50 to work and will enhance the utility of other
systems that use the sockettransportservice.
win9x provides a small number of sockets (100), so we just leave the
limits unchanged there out of conservatism.
--HG--
extra : rebase_source : 9d7a4b5a9112e17144fb510e3d8eb188919e5bf4
A force-reload now clears persistent connections to the server related
to the force-reloaded resource. This will allow renogitation of DNS or
server load balancing.
--HG--
extra : rebase_source : 5c55e90ea64039b9cdc0a2d85a51086d2b1d40df
We now store two independent locations for an omni.jar, allowing GRE/XRE and
XUL application to each have their own omni.jar. And since xulrunner setups
are very independent from the XUL applications, we implement support for both
omni.jar and non omni.jar cases in the same runtime, with the side effect of
allowing to switch from one to the other manually without rebuilding the
binaries.
We let the mozilla::Omnijar API handle both cases, so that callers don't need
too much work to support them.
We also make the preferences service load the same set of preferences in all
the various cases (unified vs. separate, omni.jar vs. no omni.jar).
The child process launcher for IPC is modified to pass the base directories
needed for the mozilla::Omnijar API initialization in the child process.
Finally, the startupcache file name canonicalization is modified to separate
APP and GRE resources.
indeed without nagle disabled (which is what nodelay does) a pipeline
can be interrupted for a full rtt waiting for an ack, which defeats
the purpose of pipelining. Other parts of FF can just coalesce the
writes internally, but sometimes that is not possible for pipelines
which may not have an opportunity for a pipelined request when the
"runt" packet goes out, but discovers such an opportunity a long time
before the rtt is expired. In other places (such as long posts) we do
a fine job of coalescing already, so this should not result in any
extra packets on the wire, just improved latency.
--HG--
extra : rebase_source : f83f5b080a3da5875abaee001e8af18503fcff38
don't send the keep-alive request header. It is redundant to
connection: keep-alive and we don't send the right syntax anyhow.
--HG--
extra : rebase_source : c6d9cb95d2d1cac30bc718884eb3b909db0d6a43
