Goal:
Saturate network I/O when downloading.
We didn't tune isolateserver much about large files download performance, so
there is likely low hanging fruits to be determined by profiling.
Proposed implementations:
These are all to be done entirely client side:
- When large files are downloaded, limit the number of parallel streams to a
few. The idea is that we don't want to starve large downloads too much so their
TCP connection get throttled, creating laggards.
- When small files are downloaded, increases connection parallelism to reduce
the negative effect of HTTP latency overhead (even when taking in account TCP
connection reuse) because of the latency of HTTP 302's that isolate server is
just too happy to reply with.
- When only large files remain, start chunked download.
Original issue reported on code.google.com by maruel@chromium.org on 15 Aug 2014 at 5:31
Original issue reported on code.google.com by
maruel@chromium.orgon 15 Aug 2014 at 5:31