Deallocate fields that are unused after a connection is established |
|||
Issue descriptionThere are some fields in some objects that we only use while establishing a connection, and no longer use afterwards. We could clear these once connections are established (HttpProxySockets keep around a lot of response metadata for instance, and HttpStreamParser doesn’t seem to free its buffers for sending a request until it’s destroyed. Not sure about SSL sockets, or H2/QUIC sessions).
,
Jul 13 2016
,
Jul 13 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/a4fc935d2fde3df3a7bb7d451b4a58f048ee225f commit a4fc935d2fde3df3a7bb7d451b4a58f048ee225f Author: maksim.sisov <maksim.sisov@intel.com> Date: Wed Jul 13 12:27:23 2016 [HttpStreamParser]: Flush buffers as soon as request is sent. This cl aims on reducing amount of memory used. Once request is sent, flush buffers to free allocated memory. BUG= 622073 Review-Url: https://codereview.chromium.org/2130733003 Cr-Commit-Position: refs/heads/master@{#405128} [modify] https://crrev.com/a4fc935d2fde3df3a7bb7d451b4a58f048ee225f/net/http/http_stream_parser.cc [modify] https://crrev.com/a4fc935d2fde3df3a7bb7d451b4a58f048ee225f/net/http/http_stream_parser.h
,
Jul 18 2016
I am trying to estimate how much memory we release here. I see that there are two ~16KB buffers being released + size of headers. Any idea how big the headers are? How many of HttpStreamParser objects can we have at one time?
,
Jul 18 2016
Their size varies a lot, depending on cookies. No idea how many are around at once. Worth noting the typical trace has 0 of them alive (Requests using an HTTP proxy would have more). On slow connections, they would also be around longer.
,
Jul 18 2016
And not to harp too much on it, but the reason why I thought we might want some sort of histograms before doing anything is just that issue. Sure, this saves memory, but no one has any idea how much, or if this particular change makes any effective improvement at all. |
|||
►
Sign in to add a comment |
|||
Comment 1 by bugdroid1@chromium.org
, Jul 13 2016