I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks. Bill Murphy University of Texas Health Science Center - Houston
Adjust your TCP window size. -----Original Message----- From: "Murphy, William" <William.Murphy@uth.tmc.edu> Date: Thu, 29 Apr 2010 10:53:01 To: nanog@nanog.org<nanog@nanog.org> Subject: Edu versus Speakeasy Speedtest I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks. Bill Murphy University of Texas Health Science Center - Houston
All the new OS's (IE Windows7) automatically adjust TCP window size. Personally I've never found those website speed test to be that accurate on fast connections (over 15Mbps full duplex). The only way to really confirm bandwidth is by running IPERF. Robert Glover wrote:
Adjust your TCP window size.
-----Original Message----- From: "Murphy, William" <William.Murphy@uth.tmc.edu> Date: Thu, 29 Apr 2010 10:53:01 To: nanog@nanog.org<nanog@nanog.org> Subject: Edu versus Speakeasy Speedtest
I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks.
Bill Murphy
University of Texas Health Science Center - Houston
Agreed. Most of the sites are not accurate for large bandwidth locations. Speedtest.net is flash based, however I find that slightly more accurate up to about 50-100mbit range. -----Original Message----- From: Bret Clark [mailto:bclark@spectraaccess.com] Sent: Thursday, April 29, 2010 10:05 AM To: nanog@nanog.org Subject: Re: Edu versus Speakeasy Speedtest All the new OS's (IE Windows7) automatically adjust TCP window size. Personally I've never found those website speed test to be that accurate on fast connections (over 15Mbps full duplex). The only way to really confirm bandwidth is by running IPERF. Robert Glover wrote:
Adjust your TCP window size.
-----Original Message----- From: "Murphy, William" <William.Murphy@uth.tmc.edu> Date: Thu, 29 Apr 2010 10:53:01 To: nanog@nanog.org<nanog@nanog.org> Subject: Edu versus Speakeasy Speedtest
I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks.
Bill Murphy
University of Texas Health Science Center - Houston
2 things. 1: http://speakeasy.net/speedtest/issues.php (See the section on inaccurate results over 20Mbps and that the test is meant for "residential broadband services") 2: Speakeasy is a commerical ISP for both residential and business users. That means it is in their best interest to encourage you to purchase their services. I have no issues with Speakeasy and have used them personally with great success in the past (great support but prices are a little high for most residential users), but why would you test one provider's service with a sales tool from another (competing) provider and expect accuracy? -Scott -----Original Message----- From: Bret Clark [mailto:bclark@spectraaccess.com] Sent: Thursday, April 29, 2010 12:05 PM To: nanog@nanog.org Subject: Re: Edu versus Speakeasy Speedtest All the new OS's (IE Windows7) automatically adjust TCP window size. Personally I've never found those website speed test to be that accurate on fast connections (over 15Mbps full duplex). The only way to really confirm bandwidth is by running IPERF. Robert Glover wrote:
Adjust your TCP window size.
-----Original Message----- From: "Murphy, William" <William.Murphy@uth.tmc.edu> Date: Thu, 29 Apr 2010 10:53:01 To: nanog@nanog.org<nanog@nanog.org> Subject: Edu versus Speakeasy Speedtest
I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks.
Bill Murphy
University of Texas Health Science Center - Houston
On Thu, Apr 29, 2010 at 9:53 AM, Murphy, William <William.Murphy@uth.tmc.edu> wrote:
I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks.
Bill Murphy
University of Texas Health Science Center - Houston
Best analogy I ever saw to teach Phd's why the net was slow: Take a vacuum cleaner with extensions. Make a set of end connectors from smaller and smaller tubes (garden hose, and straw I think they were duct taped to vacuum cleaner ends). Have the complainer try to clean up a mess with each of the ends. Ask them why it took much longer with the straw versus the regular end. For the dimwitted (eg 2-3 Phd's and various honors) elaborate that the vacuum cleaner is like your computer.. for things local and on Internet2 you get a regular hose. On going to DSlreports etc you are going at some point through a straw. [Actually i think the tube had a straw duct taped at the middle... and had things painted on it saying "What we control. What we don't control. What they control. What they don't control" ] At this point most people realized networking wasnt' the people to complain to] -- Stephen J Smoogen. “The core skill of innovators is error recovery, not failure avoidance.” Randy Nelson, President of Pixar University. "We have a strategic plan. It's called doing things."" — Herb Kelleher, founder Southwest Airlines
On Thu, 2010-04-29 at 11:48 -0600, Stephen John Smoogen wrote:
Take a vacuum cleaner with extensions. Make a set of end connectors
A "series of tubes" anyone? I'd also show them the rrd/MRTG graph at the perimeter. Be clear to them about the units. Never miss the chance to ask for more budget though. Tell them the ACL filters clog and need changing regularly, just be inventive ;) Gord -- darken room - adjust heater current so that elements glow cherry red - apply drive - tune for maximum smoke and minimum sparking
1) The capacity that a campus has into I2 or NLR is different than the BW the campus purchases from their commercial provider(s). 2) The commercial BW test sites are not optimized for speed. They do not have unlimited capacity network connections. And, they have not tuned their network stack for HS operation: notably, their OS will impose memory limits on the socket / transmit-buffer pool; so even if a receiver advertises a big window, frequently the transmitter (speed test server) will never queue enough data to fill the pipe 3) Peering capacity is not what it should be into the networks used by some of the BW test sites. On 4/29/2010 8:53 AM, Murphy, William wrote:
I work for an Edu with multi-gigabit Internet connectivity and I get questions from users saying "Why am I only getting 14Mb when I run this speed test?" I have got to believe that the various Internet speed tests (Speakeasy or dslreports) are rate limited to prevent someone from shutting them down. I am able to get 300-400Mb running from a PC inside my network to NDT servers located on Internet2, so that tells me my border and internal network is healthy. Can someone on this list shed some light regarding reliability and accuracy of these various speed tests especially for an Edu with lots'o bandwidth? Thanks.
Bill Murphy
University of Texas Health Science Center - Houston
On 2010.04.29 17:31, Robert Enger - NANOG wrote:
1) The capacity that a campus has into I2 or NLR is different than the BW the campus purchases from their commercial provider(s).
2) The commercial BW test sites are not optimized for speed. They do not have unlimited capacity network connections. And, they have not tuned their network stack for HS operation: notably, their OS will impose memory limits on the socket / transmit-buffer pool; so even if a receiver advertises a big window, frequently the transmitter (speed test server) will never queue enough data to fill the pipe
3) Peering capacity is not what it should be into the networks used by some of the BW test sites.
Your observation is disturbingly bleak... do you have a recommendation? ...perhaps a site with good bandwidth and a cluster of iperf(1) boxes available? :) Steve
On 4/30/10 3:15 AM, Steve Bertrand wrote:
Your observation is disturbingly bleak... do you have a recommendation?
...perhaps a site with good bandwidth and a cluster of iperf(1) boxes available? :)
There are better tools than a simple iperf server: http://psps.perfsonar.net/toolkit/ There are many perfsonar sites to choose from in I2/NLR land. Most gigapops and I2 networks run at least one. The problem is the Faculty^Wusers are smart, but not experienced in networking, so they buy into the marketing and eye candy of the speed dials on the Speakeasy and assorted speed testing tool sites. They see low numbers and then are difficult to convince that the devil is in the details. (which is odd, because that's what they do as scientists.) They also typically don't see the $$$$ spent on i2/nlr connections vs the $ spent on commodity internet connections. ;) Jeff
Jeff wrote:
The problem is the Faculty^Wusers are smart, but not experienced in networking, so they buy into the marketing and eye candy of the speed dials on the Speakeasy and assorted speed testing tool sites.
Not just them, we are constantly dealing with our new HS users who go to those sites then call us to complain that they are not getting the speed they are suppose to get...ugh! It's like clock work, a new circuit goes in and a few hours later we're getting the old "I went to Speakeasy...blah blah blah"! Bret
On 4/30/2010 8:49 AM, Jeff wrote:
There are better tools than a simple iperf server:
There is also http://netalyzr.icsi.berkeley.edu/ which is an excellent connectivity check, although your mileage may vary with higher-speed bandwidth testing from it. Jeff
participants (11)
-
Blake Pfankuch
-
Bret Clark
-
gordon b slater
-
Jeff
-
Jeff Kell
-
Murphy, William
-
Robert Enger - NANOG
-
Robert Glover
-
Scott Berkman
-
Stephen John Smoogen
-
Steve Bertrand