Large Upload Speed Difference between Firefox and IE 7

rickster · EVDO Junkie Joined: 27 Aug 2005 Posts: 216

Walter I only suggested that as an eye catcher. If in your note you explain your FF fan and really prefer it but explain your real life results as you have done here and are desperately seeking advise I'm 99.5% positive no one will flame you.

Please as a test also try the Opera browser http://www.opera.com/ and see what it gives you as a third party. Actually this is a great browser I use now and then for just for these types of situations; but mostly on rendering HTML. If IE is really doing something at the the socket layer that's optimizing your case Opera I'm thinking may not and should see the same results as FF. Also if you do post to Mozilla forum this will show you've done some "home work".

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

We're on the same page rickster. I understood what you meant.

And that's a good idea about Opera. I don't think I have it installed on any of my systems anymore, but I'll download it and try it as a comparison.
_________________
Reach, Connect, Celebrate, Grow, Serve.

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

Well at least this problem is very repeatable: same results on different workstation and to another site (the one recommended by ChrisM).

http://www.speakeasy.net/speedtest

Firefox:
Download Speed: 1282 kbps (160.3 KB/sec transfer rate)
Upload Speed: 206 kbps (25.8 KB/sec transfer rate)

IE7:
Download Speed: 1281 kbps (160.1 KB/sec transfer rate)
Upload Speed: 515 kbps (64.4 KB/sec transfer rate)

Tests done less than 1 minute apart.
_________________
Reach, Connect, Celebrate, Grow, Serve.

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

OK, my first Costco picture upload test is complete. Some good and some interesting (read unexplained) data:

Test setup: 7 jpg files ranging from 700 KB to a little over 3 MB in size for a total of 15.7 MBytes to upload. Using NetMeter 0.9.9.9 Beta 2 to watch the transfer speeds while it happens. Costco addin tool used for both FF2 and IE7 and told to transfer all 7 images at once.

IE7:
55 seconds total. From the NetMeter graph, looks like a bottom transfer rate around 300 kb/sec with many higher peaks up to 700 or 800 kb/sec. There were 1 or two short gaps with no data transfer.

Firefox 2:
55 seconds total. NetMeter graph looked like the lowest transfer rate was higher but the peaks were lower. There were 1 or two short gaps with no data transfer.

So with some real data... no real difference! Yay!

Of course there's a bit of an oddity with this test since 15.7 MB of jpg files in 55 seconds works out to about 2.3 Mb/sec transfer. It was definitely *not* doing that. So just what was Costco doing?

But most important, there was no difference between the two *and* the max upload speed looks to definitely be higher than 200 kb/sec.

So now the question looks to be down to just what in the world is going on with my config of Firefox and all these speedtest sites?
_________________
Reach, Connect, Celebrate, Grow, Serve.

magicjimmy · Posted: Wed Mar 14, 2007 9:32 pm Post subject:

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

Yeah, my first thought too. But here's one problem with that. I made a Zip file out of those 7 jpg files. And the zip file was 16.0 MB in size. And that was with max Zip compression on.

So unless Costco is using double secret compression... Smile

_________________
Reach, Connect, Celebrate, Grow, Serve.

ChrisM · Posted: Thu Mar 15, 2007 6:11 am Post subject:

Hi, HTTP compression (GZIP encoding) is enabled only if both the web server and web browser support HTTP 1.1. HTTP 1.1 was not enabled by default on my local installation of IE7. You can enable it in IE7 with Tools->Internet Options->Advanced->HTTP 1.1->Use HTTP 1.1. I do not know where this setting is found in FireFox.

The reason I mentioned the Apache settings for send and receive window sizes is that I have done extensive work to optimize web application performance over corporate wide-area-networks. Large (or variable sized) TCP window sizes, combined with compression yield the best results over highly latent TCP links for most applications.

The TCP window size is the result of TCP arbitration between the client (your browser running on your desktop PC) and the server (the web server running on the distant host). Both machines must support a large TCP window size in order for the communication to use large blocks for each transfer, minimizing the total number of round-trips (send, receive, acknowledge).

Our firm would not allow me to modify the company standard build of Unix to increase the TCP send and receive window sizes, so I had to search for application level settings for this. This is why I know that these settings can be set by the application, if the application is written to do so. Apache can control the TCP send and receive window sizes independantly from the operating system, using the settings I mentioned earlier in this thread. Internet Explorer does not have a setting to control these directly (at least IE6 didn't), but you can change the settings in your desktop's registry as outlined in the link I posted, also earlier in this thread, but just because there is no setting, doesn't mean that Microsoft aren't changing the values of the send and receive window sizes for the socket connections that IE creates behind the scenes anyway.

It's been about 4 years since I did this work, so some of the details are a little foggy. I believe I used the command NETSTAT to check the arbitrated windowsize of the connections at my desktop, although when I tried this on WinXP I didn't see the arbitrated window size levels, just the TCP connections, so maybe it was a different command.... On Unix I think it was "netstat -v", but I don't remember for sure. I will ask a colleague for the exact command and post his reply here. One thing I do recall is that you'll only see the window size when the connection is active, ie when you're running the test.

A tool called "IBM Page Detailer" located here: http://www.alphaworks.ibm.com/tech/pagedetailer is very good for intercepting and monitoring HTTP traffic at the workstation, and determining whether or not HTTP 1.1 and/or GZIP encoding (compression) are being used.

Why is all this relevant? Because EVDO is highly latent. The latency of the fastest local EVDO Rev A. connection is around 100ms. This is about the same as what I see on our corporate network between NYC and London (5k miles). In more local hard-wired latency would be 10 times less, around 10ms. The default TCP window size on an XP workstation is 64k. The default on most Unix servers is 8k. A typical web page is 256k. That means in the worst-case scenario, a typical web page could take 3.6 seconds to retrieve over EVDO REV A (256/8*100ms) whereas over a cable network it would only take .36 seconds (256/8*10ms), if you don't modify the default settings that is. Once you increase the window size (on both the server and the client) you will see almost the same performance in both situations.

Compression is more complex. The compression ratio depends on the content. Typical text is highly compressible, almost 100:1. Already compressed files such as .jpg images can actually get larger, and are typically excluded from compression. This is controlled at the server with settings. Compression introduces the additional latency of the time to compress and decompress the files at both ends. Obviously it makes more sense only to compress large, highly compressible web pages. This can be done through settings at the web server. For more information have a look at mod_gzip, the apache plugin for GZIP compression. http://sourceforge.net/projects/mod-gzip/

siryak · EVDO User Joined: 18 Dec 2006 Posts: 51

IMO the biggest issue here is the speed test sight you all are using. I would HIGHLY recommend using testmy.net over those, because I get MUCH more accurate results especially on upload. Just make sure you use at least the 2992kb test file.

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

OK, if 4 different speedtest sites, all with the same results, isn't good enough...

Here's the results from testmy.net:

IE7 Upload 2992kb test file: 635 Kbps or 0.6 Mbps (78 kB/s)
:::.. Upload Stats ..:::
Upload Connection is:: 635 Kbps about 0.6 Mbps (tested with 2992 kB)
Upload Speed is:: 78 kB/s
Tested From:: http://testmy.net/ (Server 1)
Test Time:: 2007/03/15 - 10:06pm
Bottom Line:: 11X faster than 56K 1MB Upload in 13.13 sec
Tested from a 2992 kB file and took 38.609 seconds to complete
Upload Diagnosis:: Awesome! 20% + : 323.33 % faster than the average for host (spcsdns.net)
U-Validation Link:: http://testmy.net/stats/id-NRWEJV0AI
User Agent:: Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1; Q312461; .NET CLR 2.0.50727; .NET CLR 3.0.04506.30) [!]

FF2 Upload 2992kb test file: 166 Kbps or 0.2 Mbps (20 kB/s)
:::.. Upload Stats ..:::
Upload Connection is:: 166 Kbps about 0.2 Mbps (tested with 2992 kB)
Upload Speed is:: 20 kB/s
Tested From:: http://testmy.net/ (Server 2)
Test Time:: 2007/03/15 - 10:15pm
Bottom Line:: 3X faster than 56K 1MB Upload in 51.2 sec
Tested from a 2992 kB file and took 147.907 seconds to complete
Upload Diagnosis:: Looks Great : 10.67 % faster than the average for host (spcsdns.net)
U-Validation Link:: http://testmy.net/stats/id-UYC3ZO1I5
User Agent:: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.2) Gecko/20070219 Firefox/2.0.0.2 [!]

So make that 5 different speedtest sites. Smile

_________________
Reach, Connect, Celebrate, Grow, Serve.

ChrisM · Posted: Fri Mar 16, 2007 1:42 am Post subject:

"Much more accurate from X test site" is totally ridiculous. Do you surf every day to the same site??????

Obviously the closest site will give you the best results, assuming it's not overloaded.

Listen to yourselves.... Use some logic. Please.

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

I hear ya ChrisM. I've done these tests enough times now to know that a different site is not going to make a difference. Based on everything I've seen with my systems, and the feedback here (especially the reports that some see this with IE7 & FF2 and some don't), I am convinced this is a real difference and that it is somewhere in the configuration of my systems: though whether Windows or FF2, I still don't know. Fortunately, the Costco test also gives evidence that although this is a real difference, it may not have any real world impact; i.e. it may just be with speedtest type sites and not come into play when I really need to upload something for real.

I played around with some more FF2 settings last night, mostly http pipeline settings and max number multiple connections stuff. Saw no difference. Also tried some maxconnections tweaks in the registry; also no impact. I think I'll still try an FTP transfer to one of my brother's computers this weekend and then post to the Mozilla forums; armed with my latest data.

Even if I fully convince myself that it's only the speedtests that are affected, I'd still like to correct this so that I can use these sites to get decently accurate info on performance.
_________________
Reach, Connect, Celebrate, Grow, Serve.

siryak · EVDO User Joined: 18 Dec 2006 Posts: 51

rickster · EVDO Junkie Joined: 27 Aug 2005 Posts: 216

Walter, did you ever try Opera ? This should take all of 15 min top to download and give a spin.

jackrodgers · EVDO Addict Joined: 23 Mar 2006 Posts: 1131

Walter · EVDO Junkie Joined: 14 Sep 2005 Posts: 307

Rickster, I did install Opera on two computers. Unfortunately, I had a few technical difficulties at the time and so don't have definitive results. What little I saw with Opera seemed to match the Firefox results more than the IE7. If I get time tonight, I'll test Opera a bit more and post back here.
_________________
Reach, Connect, Celebrate, Grow, Serve.