Chromium browser caps the TCP receive window

[j7n]

I've been playing with large TCP windows to see how fast I can get a single connection to be. And I noticed that Opera 19, which is basically a rebranded Google Chrome, scored consistently poorly in every test. It is probably important to note that the operating system is Windows XP.

To explore the issue further, I downloaded the latest vanilla Chromium 40, and I also scaled back the system's TCP window to ~512 kB so as to get it within reasonable range.

The results were the same as the other day. Trans-Atlantic speed is limited to ~450 kB/s. TestMy.net results are consistent with LeaseWeb test bins (Washington DC and the Netherlands on the screenshot).

The Receive Window has been capped to 65535 bytes without scaling, disregarding what I have configured in the OS.

Firefox 27 side by side with the other browser delivers decent speed. The speed from Washington DC increased about 9 times from 450 kB/s to 3.8 MB/s. (Firefox download manager is still dumb because it won't show neither the speed, nor the address of the download source.)

SpeedGuide Analyzer reports that my receive window is 64240 * 8 = 513920 bytes exactly as I configured it.

It is quite likely that Chromium (Opera 15+, Chrome) perform better with SPDY or under Windows 7, which can auto-tune the receive window, which I am unable to test at the moment. Somebody else could report what overseas speeds they are getting using Chrome (America to Netherlands, or to Singapore).

Chromium is definitely not the best choice for "legacy" systems running Windows 2000/XP, or maybe if Windows 7 is manually configured in the old fashion with the TCP Optimizer.

Chrome sucks..

[CA3LE]

Wow, I had no idea.

 

For the longest time Chrome worked best for me... but over the past year or so I've been having issues.  The latest... a memory leak (or something) ... every so often on my iMac I'll see little square character looking artifacts around the screen.  All blinking at the same rate, very faint... very odd.  I run through and start closing programs to find what the cause is... get to chrome, close it and BAM, problem solved.

 

I tax my other browsers just as hard and have never seen the odd behavior that chrome displays. (not just the artifacts... there's much more to it... it doesn't feel right lately) chrome used to handle everything for me but lately it's shown signs of weakness.  I have faith that the developers will make it work better.

 

That's a pretty major difference between your chrome and firefox.  Guess you know which browser you should use now.     -- from what I've seen firefox has always been a consistent performer, across all platforms.

[j7n]

Thank you, Ca3le, for the response.

This problem appears to occur only in specific conditions, which I have not yet identified. On another computer Opera 19 (Chromium 32) works fine. I've also tried a window of 256960, and it didn't change how the "broken" Chromium behaved.

Of course, reporting this on Opera forums (let alone Google's) would only lead to the recommendation to reinstall Windows, which is likely to "solve" the issue as a byproduct of resetting the offending setting along with everything else.

I have hastily misread the SpeedGuide report because of the round value of 65K. Chromium does use scaling of 2 on the base value of 32768. This doesn't change the outcome though, but looks even more bizzare. In practice, a speed of 1 to 1.5 MB/s across the continent cannot be called "slow" by any standard. But it's less than it could be.

For everone not familiar with the issue here, the TCP window defines how much unacknowledged data can be in transit, and a larger value increases throughput between geographically distant locations.