Slashdot Mirror
New Breed Of Web Accelerators Actually Work
axlrosen writes "Web accelerators first came around years ago, and they didn't live up to the hype. Now TV commercials are advertising accelerators that speed up your dial-up connection by up to 5 times, they say. AOL and EarthLink throw them in for free; some ISPs charge a monthly fee. Tests by
PC World, PC Magazine and CNET show that they do speed up your surfing quite a bit. They work by using improved compression and caching. The downside is they don't help streaming video or audio." And they require non-Free software on the client's end, too.
rproxy is a really interesting project, and back when I tried it over a 56K dial-up connection, it did actually work to speed things up. You sit an rproxy web cache at each end of the dial-up connection (so you need somewhere to deply your custom proxy to make it work, but bear with me...) and then request web pages as usual. Each end caches the pages that pass through it, but the clever part is that when you re-request a page, the proxy at the far end (on the fast connection) can fetch the page and compare with the last copy in the cache. Then it transmits only the differences using the rsync algorithm. Unforunately it's not being actively developed any more given the increasing availability of high-bandwidth connections, and the decreasing fraction of web traffic that is suitable for delta-compression. Shame, since it did seem to be a real "web accelerator" without any of the illusory techniques used by the garish banner-ad accelerators.
Matthew @ Bytemark Hosting
"Web accelerators"...You mean highly-advanced technology like mod-gzip?
Sounds pretty much like that... Which Apache already supports, and the major browsers already support, making something like this redundant.
Moreover, dialup modems already use a fairly high level of compression at the hardware layer. While not exactly "gzip -9" quality, you can only realistically squeeze another 10% out of those streams no matter how much CPU power you devote to the task.
Others have mentioned image recompression, which has traditionally used VERY poor implementations, nothing more than converting everything to a low quality JPEG. I would point out that a more intelligent approach to image compression could yield a 2-3x savings without noticeable loss of quality (smoothing undifferentiated backgrounds, stripping headers, drop the quality a tad (ie, to 75-85%, not the 20-40% AOL tried to pass off as acceptible), downgrading the subsampling on anything better than 2:2, etc). But no, not a 5x savings.
Running Squid with a 256mb ram disk cache is all the speedup we need, and it does so without altering the data being fed from upstream.
"Only in their dreams can men truly be free 'twas always thus, and always thus will be."
--Tom Schulman
why is it that downloading a 1meg file full of ZEROS still takes ages on a modem?
Because v.42bis has a maximum compression ratio of 4:1 (MNP5 only does 2:1).
Now, for a file of all zeros, hey, I agree, you can do a lot better. So, how often do you download files containing nothing but zero? For a typical text file, you might get better than 90% with gzip (while still only 75% from v.42bis). But from binary content? very rarely better than 50%.
In any case, web content consists of five basic types of information - Text, graphics, sound, multimedia (flash, MPEGs, AVIs, etc), and already-Zipped packages.
Of those, only the first benefits from any lossless method, and only the second really leaves any room for saving bits via lossy compression without horrible loss of quality at the same time. (Some of the fourth type could also possibly endure lossy compression, but takes far too long to recompress on-the-fly).
Unfortunately, text comprises the least bothersome (in terms of relative size) of all of those major types of web content.
Don't get me wrong, I fully encourage people to turn mod_gzip on in their Apache installs. But When a company hawks its product with claims that simply cannot occur in a normal webbrowsing situation, I have to call foul.
I see only two situations whereby they could claim 5:1 compression - Either VERY text-heavy material, such as reading something from Project Gutenberg, or they strip every possible non-critical image from a page. I already do the latter via my hosts file and a paranoid userContent.css, so what does that leave?
Hope you only like reading text, in which case, have you ever heard of "Gopher"?