dslreports logo
    All Forums Hot Topics Gallery


how-to block ads

Search Topic:
share rss forum feed


Wichita, KS
reply to nickfie

Re: [Info] IE8 Unleashed on Windows 7

said by nickfie:

Microsoft has not yet published this. Their internal documentation was incomplete yesterday. They did not object when I proposed sharing on this forum. In fact, last night my support engineer was eager to go home and try this on his home PC - must be a member of the "I hate to wait" club

Microsoft Support provided the RegEdit in response to a support case I opened when my packet captures showed that IE7 & IE8 are throttled to 256 KByte windowsize in Vista & Windows 7. This limit holds regardless of how you configure autotuning.

The limit applies only to HTTP, so FTP and other protocols get full autotuning benefit. DrTCP doesn't show the problem because it opens its own socket.

This means IE on Vista or Windows 7 without the RegEdit is outperformed by:

- IE on tuned XP with TCP windowsize at 512 KByte or larger. My employer uses 1 MByte for better global performance.

- Any browser on Macintosh OS X Leopard or Snow Leopard. That's when Apple saw the light - big time - about TCP tuning.

- Google Chrome on Vista or Windows 7. Google apparently saw the problem and found a solution late 2008 / early 2009.

This RegEdit changes the default performance so IE on Windows 7 is competitive. Now my employer can consider upgrading from XP without making the business work slower.
My Linux box does provided largely window size buffer support also for better global performance that been there for all years as it was been provided for, However, I set it for both minimum is 64KB and maximum is 512MB buffer that is optional for the future of generation of physical network hardware upgrades. Microsoft have learned the lessons from Linux threat
Professional Linux environmental blows microsoft windows out of the water.


Wynnewood, PA
I agree - Linux has had auto-tuning for years.

Consider raising your upper limit above 512 KByte - it is too low in many situations, and the TCP stack will only use as much memory as it needs. In our company the standard upper limit for Linux servers is several MBytes to support 20 Mbps Asia-US links. Even higher limits are required for good throughput on regional GigE links.

Also, choose your congestion algorithm carefully. A good one will provide rapid recovery from intermittent lost packets. I haven't worked in this area in a few years, but I recall that BIC was good on newer kernels & Westwood on older ones.


Wichita, KS

1 edit
I said that I'd set maximum transfer buffer is 512MB that is completely optional for the FUTURE of generation of network hardware upgrades in a case while that I set 64KB for default buffer level as normally and on side of maximum of buffer is 512MB that I set that is totally completely optional in a case when I making network hardware upgrades then that will be sufficient enough to handle a load of bandwidth and high latency of the network connection. And I use scaling TCP congestion control provider.
Professional Linux environmental blows microsoft windows out of the water.