This is something that I have consistantly noticed over years of using the internet. Namely, that whenever I download a large file, the transfer speed always falls off over time. For example, right now I am downloading a 1.3 gigabyte program ( beta test), and the transfer speed has fallen off about 40% since I started. It stared at about 120kbps, and is now down to 70kbps and falling. ( with 80% done)
This isn’t far far the only time I’ve noticed this, nor does the type of internet connection seem to matter. I’ve noticed this on Dial-up, DSL, and cable ( which I’m on now).
(At least partly)Because while you are telling the dialog box where to put the file and what to call it, it is starting to download in the background, but (apparently) the measurement of elapsed time for the download seems to start when you click OK, so to begin with, you have downloaded a large chunk of file in next to no time, but the regular performance of the rest of the download gradually pulls down the average.
For high-speed connections, ISPs put a speed cap on your link. Since it is averaged out over time, if you have not been immediately downloading something, it takes a bit before the bytes downloaded raises the average enough to kick the cap in.
Note that the way the math of the averaging is done, it makes it appear that the speed continues to go down. Really, just the average speed is going down, the momentary rate stays fixed.
Note that the other explanation offered by Mangetout works both at your end (with a well written browser) and with your ISP. Many fetch ahead links on pages customers have browsed in order to (appear to) speed up their service.
I have the same problem with my dial up connection. To begin with, my 56k rarely connects at 20k. Then the download time always crawls upwards. is it different for 56k and other dialups? (money prevents upgrading)
I fail to see how my computer could fetch ahead enough data to make any difference at all in a 1.3 gigabyte download. Certainly not enough to make a almost 50% differnce by the time the download was completed. The time delay between starting the dowload and finding a place for it is only a few seconds, while it took nearly 6 hours to download the entire file. This is far far less than 1% of the total time, and total information. That explaination makes it sound like the computer downloaded 600MB or so of information in the 10 seconds it took me to specify a file folder. Then took almost 6 hours to get the other 700, all while slowly changing the shown rate of download.
I’ve seen this in the past, but in my case it was just one limited-bandwidth, high-demand personal site that had archives of Dr. Demento shows (it’s no longer available). For the same reasons as Thaumaturge, I’m not completely satisfied with the “fetch ahead” explanation.
I always thought that ISPs gave priority to data requests that were recent. For instance, if you are surfing the web and want to load a page up quickly and then spend a few minutes looking at it then you would be given priority over the guy who is going to be downloading for the next 24 hours. This would allocate the speed to where it was most needed.
I must confess that I didn’t notice the size of the download the first time you mentioned it Thaumaturge - can you tell me if it was the first time you had attempted to download that particular file? - if not, then it is possible that (apparent)increased initial performance was brought about by a previously downloaded part of the file being retrieved from cache.
Another possibility might be that the Windows download dialog has some sort of math or accuracy problem with large files.
On links that use compression (such as dialup, or downloads from web servers using mod-gzip) the first part of the file will download significantly faster than the rest, since it is usually an uncompressed loader or header. For example, if you download a 1MB program on dialup, the first 50K or so is going to be a decompression and installation front-end for the other 950K. Being uncompressed, you can get 2:1 or better speed improvements.
Aside from this, there is the issue of caching the first few K of a file, as well as the fact that, over time, additional bandwidth usage will affect the average download rate.
Internet Explorer caches recently downloaded files. 600 MB in 10 seconds is 60MB per second… which is basically impossible over the Internet. Some portion of the file had to exist before you began the download.
No, I had never downloaded the file before, nor had any version of it existed at any time on my computer before. It was new software being sent to be for beta testing. I dowloaded it off of an FTP site, but I have noticed this behavior numerous times in many different situations as I noted in my OP.
“Another possibility might be that the Windows download dialog has some sort of math or accuracy problem with large files.”
This is probably it, though there has to be someone out there who knows for sure.