Slashdot Mirror
3Com's "Gamer" Modem Pings Faster?
An anonymous reader pointed us to 3Coms Gamer Modem:
they claim faster ping times and better online play. I'm more than a little skeptical here, does anyone have more info?
← Back to Stories (view on slashdot.org)
In a 1998 survey of various ISPs, it was noted that 57.9% of computers ping to stay online. Problem was, it was wasting too much processing time on the server, causing some ISPs to have special servers just to respond to pings. That was all dandy until some kids decided to play "ping-pong" with the servers in a synchronized attack, and brought the ISP to its knees.
Well, one Admin decided it was time to get back. He logged three days worth of pings from the most used accounts, and started responding to pings that didn't exist. The hack that the guys wrote to ping the computer in the first place was not ready for it, and took down the computer.
The Admin spread the news, and it caught like wildfire. Soon there were many variations of the program and each added it's own flavor.
Then it hit. One guy realized that the amount of time responding to valid pings was double what it should be. But if the ping response was sent at the same time as the ping, it would cut the time down to half.
Working on this theory, and collecting average ping reports. Mike Roe Chip, Network Administrator for ISP Communications, designed a protocol in which the ping responses are sent out at the same time as the ping itself. It's is correct 99.99999 percent of the time, according to his Pentium(TM) based calculations. To make up for any incorrect responses, it sends out a Ping Response Cancel Packet. The new protocol is called DCPP (Detect and Correct Ping Protocol), based on APT (Advanced Pinged Technology), and is coming soon to servers in your area.
Have you read my journal today?
Actually, it is much more sophisticated then simply detectecting dead frequencies, and moving the signal around, and in fact it was thought of sooner, and has been done for a long time.
Think of the signal coming in as a spectrum (like winamp shows you in it's default mode). This is basically a bar graph, with frequency along the bottom axis and power on the side axis. When somebody starts jamming on their bass guitar, you get bumps in the low frequency end. When they start playing the flute, you get bumps in the high frequency end.
As a side note, I believe mp3 encoding takes advantage of this concept to achieve it's high compression rates. It (metaphorically) saves the height of the bands, and reproduces those on playback. The higher the number of (more narrow) bands, and the more accurately you measure their height, the better your sound reproduction.
Any transmission medium will distort this spectrum to some degree. If you look at the spectrum on the sending end, and the spectrum on the receiving end, you will see it changed.
Your average phone line has a pretty narrow spectrum that it can transmit. I think it ranges from about 500hz at the low end, to 3500 hz at the top end. A normal CD reproduces sounds from 20 hz to 20000 hz. Unless you are a pre-pubescent female, you likely can't hear much above 15000 hz. ( hz=Hertz, cycles per second).
So anyway, if you want to get more bandwidth (lower lows and higher highs getting crammed through a phone line), there is a neat trick to doing it (which is also used by Bose on several lines of their speakers with great results).
1) Send a known signal through your transmission medium.
2) Receive that signal, and compare it to what you sent.
3) Before you send your next signal, pre-distort it, so that when your transmission medium reshapes it, it ends up at exactly the shape you wanted in the first place.
It's kind of like buying jeans that are not pre-shrunk... by them long, so that when they shrink after being washed they end up the size you wanted.
Simple, huh?
The reason this is becomming more and more common, and that data rates are so amazingly high for such lousy transmission mediums like phone lines, is that heavy duty signal processors are just now becomming affordable enough to embed in consumer devices. It's been around for quite a while, you just could not afford it.
Signal processors (DSP = Digital Signal Processors) are simple computers that have very limited functionality but do their job blindingly fast. This functionality is now to the point where it can be embedded in a single chip, and sold for a few bucks, but the engineering that goes into these things is staggering.
There are other methods of error correction that were necessary to get data rates up the 56k speeds we now see, but they are pretty complicated.
Bill Kilgallon
Mathematically impossible requirements are technically not against policy.
So the 3com article says that an independent testing laboratory found that the game modem was 43% faster than other modems when connected to a 3com Total Control server. Here is the independent testing lab, and it's documents describing their test setup/testing methodology:
Henderson Labs
I have submitted a request to henderson labs to make the 3com test results publicly available. Please don't spam these people, I will post the results if/when I get them.
Can a modem be made better for gamers? You bet. Same way a protocol can be made better for gamers. It's all about LATENCY...
Why do we use UDP instead of TCP for gaming connections? Because we don't care about checksumming (checking for correct data) or compression (more data faster on average, but higher LATENCY), or RETRANSMISSION. That's the big one folks, retransmission. In a game, if you've lost a packet, you don't want it retransmitted anyway, because it probably is carrying the co-ordinates of a guy who's already killed you 10 seconds ago anyway. If you're tranferring a file, you definitely want that packet. (:
Real-time applications (read: games) need low-latency connections above all else, even an unreliable connection. Normal I/O needs reliable SAR (segmentation and reassembly). Those are very different goals.
Here's the REAL question: Does this modem do anything you can't reconfigure a normal modem to do using the hayes command set? Does it turn off some retransmission features that we can't normally turn off ourselves?
Or is it just a different set of default flags bundled with some lame games for l33t hax0rs in time for christmas?
3com, here is your chance to speak up. Engineering btw, not "product management".
-I am Jack's identity crisis, in full gear.
--
What happens when you outlaw guns
1) Compression in most modern modems is done using the V.42 specification. V.42 is actually very complex protocol which specifies frames that can be compressed. So you have to buffer data before compressing and sending. Error correction is done at the 'frame' level. There is a minimum latency to fill up the buffers before transmitting data.
Bottom line: If you disable V.42 you can subtract some milisseconds of your typical latency time. The connection is less reliable and have less bandwidth available, but its better for games.
2) The old fashioned UART interface is very slow. Even a 16550 uses a very small buffer for today's standards. This is one of the reasons to use a USB modem for better gaming experience. If this is a Winmodem, thay may have opted for a faster interface to transfer bytes from the CPU to the modem. Maybe they are using something like an ethernet interface, with larger packet buffers.
Bottom line: its possible to have a faster modem by using a faster interface instead of a plain UART.
Hi!
I'm an engineer in the 3Com modems group. The Game Modem is optimized for small packet sizes. Most games use tiny packets to indicate stuff like rudder position and etc, so this works better. Of course, short-packet pings fall into this category, too. Also, we ship it with some quite nice games in the box.
And, yes, it works just fine for the traditional things you use a modem for.
Happy Halloween!
Dog is my co-pilot.
-- Support Ometz le-Serev.
In fact, Henderson Communications Laboratories, an independent testing lab, concluded that the 3Com Internet Gaming modem achieved up to 43 percent faster weighted average ping times than five other competitor modems tested this October when calling to 3Com Total Control® or Ascend Max server equipment.
As with all 3Com U.S. Robotics V.90 56K modems, the Internet Gaming modem includes x2(TM) technology to ensure backward compatibility with x2-capable modems and server equipment.
So it only works with their hardware on both ends, but is backward compatible. Naturally. Also:
The Internet Gaming modem also takes advantage of 3Com's exclusive line probing technology. This feature allows 3Com modems to dynamically create a signal pattern that optimizes throughput for that particular line, ultimately leading to higher levels of performance than other V.90 designs.
Line Probing? Heheheheh good one Beavis...
It probably just shifts frequencies and so on to test which signals are least likely to have interference due to line static. Then it uses those.
Which is a bit interesting.. Makes you wonder why the hell something like this wasn't invented years ago. I mean, if the modems can sort out speeds and so forth between them, why not freq. as well? Would saved me a hell of a lot of redials and restarted downloads...
---
- Give a man a fire and he's warm for a day, but set him on fire and he's warm for the rest of his life.
Consider for a moment that compression has a time delay factor intrinsic within it--the longer one waits, the more redundant data can be filtered out of a transmission block.
Modems, by default, execute Run Length Encoding(RLE) algorithms, which if I remember correctly are statements along the lines of "Here's a string of 64 0's" instead of literally sending the stream of 0's.
Of course, to *know* one is sending 64 0's, one has to operate on a 64 byte delay. So a key strategy for reducing lag is actually sending those 64 0's live rather than waiting the delay period.
This isn't really a bad idea--waiting for the delay period when highly tuned networking applications which would *never* send such a "low entropy"(translation: almost devoid of unique information content) string is foolish.
Now, disabling modem compression is a well known tactic for decreasing ping time, and tools have been out to reconfigure Windows to do thus for years. That doesn't necessarily mean, however, that all these modems have different is a new driver disk.
56K modem technology has been described as the biggest technical hack the industry has ever seen, and I'm inclined to agree. That it works at all is near-miraculous(although actually it doesn't really go as fast as advertised, thus 3Com's upcoming $5 coupon slap on the wrists for claiming net connections would be twice as fast).
Intrinisic in the protocol are error-checking codes. Error checking *also* introduces delays, as you need to wait for the data to come in before you can sum it. Reduce your error checking, or decrease the check interval, or tweak in any number of hacks, and lag can decrease.
Also intrinsic is connection recovery--by speeding this up, making this more effective, or both, 3Com gets an edge. That there appears to be specific functionality reserved for specific ISP hardware leads me to suspect there's off-standard code being put into use.
This isn't necessarily bad.
I'd be interested in more technical documentation as to what they've done--anyone from 3Com got a real link for the rest of us?
Yours Truly,
Dan Kaminsky
DoxPara Research
http://www.doxpara.com
I recall Carmack making a .plan update or two while he he was talking about latency while creating the first version of Quakeworld. He stated a few changes which could be done in order to decrease latency.
:)
What your modem does, by default, is collect all data into a buffer, and then send it out all at once. The larger the buffer, the more efficiently compression can be performed upon it. All of the old Doom init strings that came with the game turned off your compression... and your error correction for that matter. (Wasn't it ATL0&M0&N0&B0 for a US Robotics?)
A common misconception is that some of the new midspeed connects such as ADSL or Cable give low pings because they have a lot of bandwidth. Rather, it's the lack of PPP. PPP adds about 140ms onto the ping time, because of the way it's designed: SLIP looks like it was written on the back of a paper napkin. PPP was.
Having strikingly low bandwidth causes for dropped packets, which is a very high possibility, especially since most (not all) packets in Quake games are UDP. However, other than this exception, bandwidth is secondary to latency when playing Quake.
That's why there isn't a notable latency difference between 28.8 and 56k. It's still PPP with error correction.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)