Speed Tests All Over Map; Is Tweaking Viable?

[fasterjay]

We recently changed from AT&T DSL to Comcast business cable. The package is for up to 50 Mbps up and 10 Mbps down. We have a NetGear N600 (also called WNDR1300) router, which I know for sure is configured properly.

Comcast came back after the new installation of service to this building and the guys really did a Herculean effort of tweaking the outdoor amplifier, check the lines, modem, etc. I'm very confident everything is fine now.

Speed tests are showing all over the map and I have no idea what's what. I can understand slower speeds at longer distances but the variations I'm seeing are incredibly wide from test site to test site and when using FTP software.

All test results shown below are "per second". We're located in South Florida.

The following tests were done our XP machine:

testmy.net shows about 11.2 MB down and 8.2 MB up.

speedtest.net shows 52 MB down and 12 MB up.

speedtest.comcast.net also shows 52 MB down and 12 MB up if we use their server that's close to us in South Florida, or if we use the one in Atlanta. Using their server in Boston shows 30 MB down and 11 MB. Using their server in California shows 14 MB down and 11 MB up.

www.dslreports.com is really varied. Using their Flash tests via Comcast in Washington DC we get about 9.8 MB down and 3.8 MB up. Using their Flash via SpeakEasy in NY we get about the same. Using their Flash via LinkLine in Los Angeles we get about 5.7 MB down and about 1.2 MB up with a warning that the upload speed may be inaccurate due to compression being detected.

www.dslreports.com using their Java test shows 8 MB down and 3.6 MB up, and using LinkLine in California it shows 5.6 MB down and 2.8 MB up.

Then, there's my testing of actual file transfers.

On downloading a trial version of software from the Adobe website, the Adobe download window showed a download speed of about 6.3 MB.

And on using my FTP software, which is a five-year-old version of WS_FTP Pro, uploading and downloading large (100 MB and up) zip files to and from my graphic designer's server in Boston, and to and from our web host's server (a top-rated firm located in the middle of Florida), we show a range of about 1.4 MB down (from our web server) to 550 "KB" down (from our graphics designer), and generally about 780 "KB" up across the board.

Now, the same tests on our Windows 7 machine:

testmy.net shows about 42.2 MB down and 11 MB up.

speedtest.net shows 52 MB down and 12 MB up.

speedtest.comcast.net shows 52 MB down and 12 MB up if we use their server that's close to us in South Florida. If we use the one in Atlanta it shows 52 MB down and 9 MB up. Using their server in Boston shows 30 MB down and 01 MB up. Using their server in California shows 52 MB down and 6 MB up.

On downloading a trial version of software from the Adobe website, the Adobe download window showed a download speed of about 5.2 MB. (This surprised me, being slower than the XP machine, so I repeated it and got the same thing again.)

Using the FTP program was roughly similar to on the XP machine. The downloads showed about 5% faster than the XP machine, and the uploads were about comparable to the XP machine.

I'm guessing that the FTP has some inherent limitation, either because it's FTP, or because of settings I need to tweak somewhere, or because of inherent limitations at the servers that I'm transferring files from and to.

The only tweak I see available for the FTP software is to have it do "multipart" transfers, breaking the file into parts and establishing multiple connections "if" the server in question allows. I don't know if the servers will support it or not. It seems I'll need to install the newer version of the software to try that. (I'm thinking of changing to SmartFTP, though it seems it only provides more security, which doesn't seem to be an issue for us.)

I've been reading a bit about tweaking Windows XP. DSL Reports recommends DRTCP for adjusting some settings. I have no idea how safe or helpful that might be. I haven't tried it. I feel I better look before I leap. It says that Windows 7 is already optimized, or, at least, one of the main tweaking settings is.

I really need speed for transferring large video files with our graphic designer. We've done 10- to 20-hour transfers in the past. I have a bad feeling that what we're seeing here is that the bottleneck is on his end.

We also need speed for uploading to our server, though that's usually FLV files which tend to be smaller.

I absolutely can't risk destabilizing my XP system with tweaks since it's mission-critical with a lot hanging in the balance.

Any suggestions would be enormously appreciated!

Thanks!

[CA3LE]

Well, looks like speedtest.net and speedtest.comcast.net aren't detecting your issue. This is pretty common. This is most likely because the tests here take TCP Windowing into account.

DRTCP (from DSLReports.com) or TCP Optimizer (from Speedguide.net) should correct the problem. Many times in cases like this the receive window is set too low. The changes made by these programs are relatively safe and I know that TCP Optimizer has windows default values you can resort to if you want to reverse your changes. Also, you can back up your registry if you're worried about the changes having an ill effect.

XP is notorious for it's bad TCP/IP values out of the box. Chances are that this fix should be quite easy for you. Simply apply the values, restart and re-test.

[fasterjay]

Well, looks like speedtest.net and speedtest.comcast.net aren't detecting your issue. This is pretty common. This is most likely because the tests here take TCP Windowing into account.

DRTCP (from DSLReports.com) or TCP Optimizer (from Speedguide.net) should correct the problem. Many times in cases like this the receive window is set too low. The changes made by these programs are relatively safe and I know that TCP Optimizer has windows default values you can resort to if you want to reverse your changes. Also, you can back up your registry if you're worried about the changes having an ill effect.

XP is notorious for it's bad TCP/IP values out of the box. Chances are that this fix should be quite easy for you. Simply apply the values, restart and re-test.

I appreciate it very much.

There's one reason I haven't been thinking much in terms of using either of those utilities, which is the fact that DSL Reports says the tweaks only apply to XP, not Vista or Windows 7. Or maybe it's really only saying that about "some" of the tweaks. On its tweak-test page it says:

WINDOWS VISTA USERS: Vista TCP tuning (tweaking) is NOT advisable. The Vista TCP stack does a reasonable job of tuning the receive window dynamically, there are no known registry tweaks that would optimize Vista better than its default configuration.

And I had seen somewhere on their site a comment saying that the same or similar reasoning applies to Windows 7.

So, to sort of compare what I might get if I did the tweaking on my XP machine, I ran all the same speed and real-world transfer tests on my XP machine and on my Windows 7 machine, and saw very little difference (and in some cases the XP machine was a bit faster).

Maybe I'm overlooking the difference between TCP adjustments and RWIN adjustments, or something like that?

When I run the DSL Reports tweak test, here's what I get.

What do you think?

Thanks very much!

Jay

[fasterjay]

I should add to my previous post the facts that I have now learned the difference between Mbps and MB/s and should have been saying Mbps in my OP, and that I'm pretty convinced the problem is mainly a bottleneck with Comcast. Their tech who did the local amplifier adjustments was candidly talking to his co-worker and me about how much Comcast is pushing far more users through the given node than it was ever intended to handle, and said that if I was three miles west I'd be getting far better speeds.

So, do you think I "may" have much to gain in trying tweaking anyway?

Thanks again.

Jay

[fasterjay]

To add to my previous post...

As much as I'm focusing on congestion within the local Comcast system, I realized that something I'm seeing doesn't necessarily jive with that theory, which is the fact that every speed test I do, on both of my computers, is showing remarkably similar results -- within 10% variation -- for the given test method and the given OS/machine. I've done this over a number of days, more than half-a-dozen times, spanning business hours, late nights, and weekends.

It seems to me that if it was a bottleneck issue with traffic, the speeds would be varying far more widely than just a 10% range for the same method of test.

And when I say test I'm including actual file transfers to and from different locations.

The only exception I've seen is about a 35% change with uploading files via FTP to my graphics person's server, and when I hit the high end it was essentially identical to the speed I'm seeing when uploading via FTP to the FutureQuest server. So, really, even that has shown consistency for the most part.