Date: Thu, 12 May 2005 08:10:43 -0500 (CDT) From: Adam Maloney <adam@whee.org> To: freebsd-hackers@freebsd.org Subject: Re: upload speed test problem Message-ID: <Pine.GSO.4.60.0505120749190.2099@titan> In-Reply-To: <6.2.1.2.2.20050512180519.03403540@202.179.0.80> References: <6.2.1.2.2.20050512180519.03403540@202.179.0.80>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 12 May 2005, Ganbold wrote: > Result is unbelievable, it is something like 500kbps for 56kbps Dial-Up > connection, which is completely wrong. It looks like the form data that the client is POSTing is all "a", correct? The modem is probably compressing this data. I wrote a speed test in perl a number of years ago to do both upload and download testing. Here are some things I found, maybe they will help you: You must use data that doesn't compress well, or, as you've seen, dial-up modems will compress it and report speeds much higher than is possible. It is better to pre-generate the random data, rather than try and read from /dev/random on the fly. I realized that by having my script suck 5 or 10 megabytes of randomness out of /dev/random every time it ran, I was accelerating the heat-death of the universe. You don't want to cause the heat-death of the universe either, do you? (Actually, either pre-generate random files, or read from /dev/urandom, which doesn't block when it runs out of randomness) IE Sucks. The way my script worked was, I had a form page that the user selected the file size to test with, and hit Submit. This submitted to my CGI that generated a new HTML form with a hidden input field containing the random data of the size they selected. It also contained a hidden field with a timestamp of when the page began to execute. I used javascript to automatically submit the form when the page load completed. Mozilla would submit this form as soon as page load completed. For some odd reason, IE would wait a couple of seconds before submitting the data. So I had to (oh this is so ugly, I don't want to say it), define an IE fudge factor, and subtract a couple of seconds from the upload time if the client was IE. I also realized that I had to add a META tag to cause the pages not to be cached. Furthermore, I had to add something to look at the client's HTTP headers and look for signs of an HTTP proxy (proxies usually add a header or two, depending on how they're configured). If a proxy was detected, I could either spit out a warning to the user that the speeds reported could be inaccurate, or I could simply refuse to continue. On the plus side, overall the test worked pretty well. HTH, Adam
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.GSO.4.60.0505120749190.2099>