Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
 ·Comcast
|  LinkSys Speed Test Results of fw 1.40.2
Of course I just notice 1.42.3 is out! 
BEFSR41 Version 1 (vintage June 2000).
F/W 1.40.2
Router is at default settings (after a loooong RESET).
Using WS_FTP on LAN to BPFTPServer on WAN (2PCs).
Actual (raw) rate measured with NetPerSec Ver 1.1
WS_FTP provides TCP data rate.
LAN machine is Win2KPro 333MHz K6 (100mbps into LinkSys LAN).
WAN machine is Win2KPro 450Mhz P3 (10mbps on LinkSys WAN).
I only use 1mbps = 1024kbps, 1kbps = 1024bps, etc. but will include DSLReport's expected reading with The Constant 2% Error.
Maximum Upload:
Raw: 6042kbps
TCP: 5174kbps (647kB/s)
(DSLR would report 5298kbps)
Maximum Download:
Raw: 8090kbps
TCP: 7464kbps (932kB/s)
(DSLR would report 7643kbps)
These did NOT affect these readings significantly (all less than 1/2%):
All options switched around on the bottom of "Filters" page (except MTU).
Enable Logging
Set WAN to static settings
Set LAN PC to static settings
Disabled LAN DHCP
Port Trigger entry (no triggering occurred)
Port Forwarding and DMZ slightly reduces upload:
Either or both, it doesn't matter, reduces maximum upload by about 3%. Download was not affected.
MTU difference measured
I reduced the MTU from default (1500) to 1362 (a low but standard PPPoE value).
Only download was degraded by 7%. I'm not sure why upstream wasn't affected but it was not.
How are the 2 boxes without the LinkSys?
I wonder if I may have some anomaly in either of my LAN cards. This is because I ran the same systems connected to each other by the LinkSys 4-port switch so both negotiates 100mbps (remember the LinkSys WAN has to run at 10mbps). The download/upload anomaly continues:
Upload approx 13mbps
Dnload approx 26mbps
All Comments Welcomed.
--
Expert Opinions: $5... I Shut Up: $10
[text was edited by author 2002-01-31 01:14:31] |
|
bbarrera
Premium,MVM
join:2000-10-23
Sacramento, CA
clubs:
 ·SureWest Internet
| Bill, last week I sent my brain spinning around "K = 1000" vs "K = 1024" debate.
Bottom line:
Datacom/telecom/wireless/clock_freq uses decimal
K = 1000
M = 1,000,000
- ftp shows 100KB/s and that equals 800kbps
- 450MHz clock is 450,000,000 Hz
- 10Mbps Ethernet is 10,000,000 bps
Computer memory uses binary
K = 1024
M = 1,048,576
- 256MB RAM is 268,435,456 bytes
After dusting off old telecom/datacom/networking text books, I found this website to clear up my confusion:
»www.speedguide.net/Cable_modems/···th.shtml
edit: I then proved it to myself by doing long ftp downloads under Win2k and Linux, and then hand calculating transfer rate. It was nearly identical to speed report of ftp program. Haven't looked at DSLR, NetPerSec or WS_FTP, so I can't comment on your environment.
edit: fixed browser link
[text was edited by author 2002-01-31 12:37:11] |
|
bbarrera
Premium,MVM
join:2000-10-23
Sacramento, CA
clubs:
·SureWest Internet
| Curiosity got me to look into DSLR speed test. I put it into verbose mode and here is one of the downloads:
Downloaded 696000bytes in 2985ms
First guess is 1865kbps
Which according to my calculator:
1,865,326.63bps = 1865kbps = 8*696000/2.985
So DSLR speed test is correctly using decimal "kbps" in calculating download/upload speeds. |
|
Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
·Comcast
| reply to bbarrera
Hi bb, yeah, it's a mess. The old datacom 56k (via RS-232) was 57344bps but flex 56k modems run off convenient clocks that easily tolerate even 3% error!
Any time I've ever used a speed utility to measure network speed it's always been 1024-based. This JAVA applet used by DSLR (and several other places) is the only exception. Check out every other online speed test that you can see it - bet it's 1024 based. I never saw an ftp speed measure that wasn't 1024-based.
BTW, that link is broke... typo? I'd like to see what it says. |
|
Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
·Comcast
| reply to bbarrera
said by bbarrera:
So DSLR speed test is correctly using decimal "kbps" in calculating download/upload speeds.
Decimal was never the norm. Heck, even my old Telocity 784k/784k SDSL modem had the exact rate as 802816bps! EarthLink DSL download is called either 1.5mbps or 1536kbps. |
|
GeekNJ
Premium
join:2000-09-23
Waldwick, NJ
| reply to Bill_MI
Bill, just so I understand... your MAX download that you show near the top of your post, means what? The MAX that you could download?
For example, I have OOL, and on the Speakeasy tests, I can get 8500kbps down. On the OOL FTP server, I get in IE, approx 1.04MB/sec speeds.
--
Have you tweaked your OOL connection? |
|
bbarrera
Premium,MVM
join:2000-10-23
Sacramento, CA
clubs:
·SureWest Internet
| reply to Bill_MI
said by Bill_MI:
Decimal was never the norm. Heck, even my old Telocity 784k/784k SDSL modem had the exact rate as 802816bps! EarthLink DSL download is called either 1.5mbps or 1536kbps.
Bill, I respectfully disagree with Telocity and EarthLink (my ISP and Speakeasy use comms definition of k -- 384k is 384,000bps). My 3Com ADSL modem shows 384,000 bps for my 384/128kbps service. I designed modems and telecom equipment in the 1980s, and back then binary 1024 was only used in reference to computer memory since computers use binary. My spectrum analyzer and oscilloscope never used binary k=1024.
I'll admit the software industry has distorted the definitions used by the comms industry. The GUI for the online speed tests were probably designed by software engineers, and they incorrectly used binary arithmetic instead of decimal arithmetic for speed.
For example, read this page from Speakeasy that clearly states 384kbps service is 384,000 bps. It also notes that your browser (written by a software engineer) will require converted speeds with the "1024 factor" -- »nyc.speakeasy.net -- doesn't mean the software industry is correct but that's what they've done.
My new bottom line:
- Comms industry normally uses decimal version of k/K/M/G
- Software industry in general confuses people by using binary version of K/M/G in reference to speed
BTW, I fixed the browser link in my other post.
Great job on testing, another thumbs up.
[text was edited by author 2002-01-31 12:46:57] |
|
bbarrera
Premium,MVM
join:2000-10-23
Sacramento, CA
clubs:
·SureWest Internet
| reply to Bill_MI
Bill,
Checkout the ftp transfer rate report in Win98, Win2k or Linux. Use the default ftp client. For example I'm on the Win98 box downstairs and just ran ftp over the Internet:
ftp:5655393 bytes received in 37.24Seconds 151.86Kbytes/sec.
Which according to my calculator:
5655393/37.24 = 151863.4 bytes/sec = 151.86Kbytes/sec
which is matches the FTP report. I've checked and verified that with OS supplied ftp programs under Win98, Win2k and Linux.
By the way, think about it. Your a comms engineer. The spec says 1Mbps. You want to look at bit stream on an oscilloscope. Easy, set your time base to 1 micro-second (the inverse of 1M) and your good to go. No messy "remember to multiple by 1024 first," just pull out the calculator, type in the bit rate, and hit the 1/x button. I think it is pretty clear that the comms industry standardized on decimal usage of k/K/M/G a long time ago, and the rise of computers has only confused matters. |
|
Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
·Comcast
| BB, I guess I'm stuck in the old school. Hey, THIS is a new one on me: »www.romulus2.com/articles/guides···es.shtml
Kibibit??? LOL! Everyone wants to put it to rest - even IEC.
I did do some reading and found many supporting (and often confusing) sites, though...
»www.yourmojo.com/about/glossary/···ary.html
»www.firmware.com/support/bios/metric.htm
»www.youthtech.com/techstuff/tech···rmsk.htm
»www.kornets.com/chobo.htm (japan)
»www.intermountain.com/internetTerms.html
»www.net-jobs.com/intranet/GK.htm
»www.kcw13.dial.pipex.com/indexij.htm
»www.iupui.edu/ithome/training/qu···dem.html
|
|
bbarrera
Premium,MVM
join:2000-10-23
Sacramento, CA
clubs:
·SureWest Internet
| said by Bill_MI:
BB, I guess I'm stuck in the old school. Hey, THIS is a new one on me: »www.romulus2.com/articles/guides···es.shtml
Kibibit??? LOL! Everyone wants to put it to rest - even IEC.
That was a great link and summary of the issue, much better than my link. You really have to sharpen the pencil and determine for each application how the transfer rate is reported. I don't use graphical front-ends to ftp except when I mistakenly click on a link with IE5.5. I've found that the old school ftp software got it right.
Reminds me of the 1980s -- people referred to "2400 baud" modems when in fact the baud rate was 600 (4 bits per symbol or baud) and the bit rate was 2400. Something only a designer (me) or purist would appreciate. |
|
Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
·Comcast
| reply to GeekNJ
said by GeekNJ:
Bill, just so I understand... your MAX download that you show near the top of your post, means what? The MAX that you could download?
Hi Geek, that's the max I can push with no ISP involved. It's a PC on both LAN and WAN to test. Now what would make your extreme top end faster? Could be my system(s) - they're not the fastest . As you approach 10mbps the system differences have to come into play - literally TCPIP stack latency. I cannot explain the up/down anomaly, yet. When I saw a direct connection showed the same up/down skew it got me leery of the results.
Hey, a good selling point to the wife to get a new motherboard... |
|
GeekNJ
Premium
join:2000-09-23
Waldwick, NJ
| reply to Bill_MI
Bill - what is your RWIN set to on each of those machines?
I have a 200MHZ Pentium hooked up to the same Linksys and it gets 6000 down off the Speakeasy test.
--
Have you tweaked your OOL connection? |
|
Bill_MI
Bill In Michigan
Premium,MVM
join:2001-01-03
Royal Oak, MI
·Comcast
| RWIN is 65535 on both. My little spreadsheet says for RTT of 10 msec I only need RWIN of 12800 to support 10mbps - so it's a little overkill. 65535 is good to 50 msec RTT at 10mbps.
You're familiar with the speed/RTT/RWIN relationship, right? I'd want a bigger RWIN for the real net - probably 4 times that if I wanted reliable 10mbps not waiting for ACKs. If I set this up again, I'll see if it does anything on this test.
Yes, I believe 6000 - I got 7643. Speakeasy uses the same applet as DSLR.
[text was edited by author 2002-01-31 15:30:33] |
|
