Slashdot Mirror
How Fast Too Slow? A Study Of Quake Pings
Jonathan Lennox writes: "Grenville Armitage has published an analysis of ping times of clients connecting to his Quake III Arena server. His conclusion is that 150 milliseconds is the limit that people find tolerable, and says this may have interesting implications for Internet Quality of Service research."
An interesting study might have been the average score for each ping, and seeing where the dropoff area was, rather than just people who quit or never joined.
Now why can't someone do this study on framerate and settle THAT argument? :P
Ping times determine the latency of a given connection. Latency is the time that it takes to get one response from a server to a client upon the client's request. Bandwidth affects latency- and is just about one of the only things you can do to impact it. Switches insert latency. Routers insert latency. Modems insert latency. Heavy traffic inserts latency. Once you get latency, you can never rid yourself of it without major changes to your infrastructure. To put this all in perspective: The average latency without heavy traffic losses for a xDSL node to the frame relay at the DSLAM is something like 5-15 msec. The average latency for a Cable node to a border router is something about 30-50 msec on an unloaded segment. The average latency for a dialup modem (of ANY speed) is 250 msec- about 1/4 of a second. The average latency for an 100-base-T adapter is 1-3 msec. The average router latency is something like 2-5 msec. The latency on frame relay can be a couple of msec if it's not loaded to hundreds of msec as it gives higher priority packets credence under load. Each one of these things add up- and fast. And these don't go away unless you rip them out for faster equipment- adding bandwidth doesn't make those parts of the latency go away. That "speed of light" ping time you referred to just went "poof" because it was ate by the intervening hardware.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Not being a Q3A player (I won't buy a copy until they remove the copy control measures), I can't speak directly to Q3A latency in general or his server in particular. However, having been a QuakeWorld addict for several years now, I have several opinions on what contributes to overall quality of game play.
First off, let me repeat as others have above that being an LPB (Low-Ping Bastard) does not assure you victory against all comers. I have been well and truly thorked by players sporting 200+ pings. I am also a decisively mediocre player.
While round-trip time to the server ("ping") is important, I find the game's rate to be just as important. rate is the amount of bandwidth your client will consume, and defaults to 2500 (bytes per second) for QuakeWorld. This limits how much game state can be updated at any given time. Lower rates mean less game state data and reduced "fluidity" of game play. So even if your ping is <20ms, a rate of 2500 gives you incredibly choppy updates. Thus, if you have a high-bandwidth connection (DSL, cable modem, OC3), the first thing you should do is crank your rate. I keep mine set at 8000, which gives me much smoother, much more fluid display. I could set my rate to 10000 since I have high-bandwidth SDSL, but I keep it down a bit as a courtesy to server operators so as not take more bandwidth from the server than I need.
Your rate setting should not exceed the total bandwidth available on your net connection. Thus, setting your rate to 6000 on a 56k modem will actually make things worse. Games with the rate setting include Quake, QuakeWorld, Quake2, Quake-3 Arena, HalfLife, CounterStrike, and all other Quake engine-derived games. Non-Quake engine-derived games also have this kind of setting, but they all call it something else.
Another big source of latency can be your ISP and their Quality Of Service metric. I used to have 416K SDSL through Best/Verio before Northpoint cratered. I have since switched to Speakeasy 1.1M SDSL, who partners with Covad. With Best/Verio, the best pings I saw were in the 20-25 ms range. Occasionally I would get an 18. When I switched to Speakeasy, I naturally assumed that the higher bandwidth would yield even lower pings. However, this has not happened. After some study with traceroute, it's my suspicion that Speakeasy's routers are either overloaded or configured sub-optimally. I posted my thumbnail analysis to Speakeasy's discussion fora (mine's the last post in the thread), but haven't received any response yet.
Another way the network can hose your gameplay is by dropping packets. Speakeasy seems a little more willing to drop the odd packet than was Best/Verio. However, Best/Verio had a router in their network that would occasionally go apesh*t and drop all packets for about 90 seconds, which is long enough to get you disconnected from any server.
Games seem particularly vulnerable to dropped packets. QuakeWorld is the only game that seems to handle this issue robustly, patiently waiting to re-sync with the server. Nearly every other network game I've played -- Quake2, Unreal Tournament, HalfLife -- will never re-synchronize, and you have to reconnect. Serious Sam is especially bad in this regard (I'm willing to cut Croteam some slack since this is their first product, but I hope they fix it soon).
BTW, most of the Quake engine-based games have a little network diagnostic tool called netgraph, which shows you a realtime scrolling graph of your network latency, including dropped packets, corrupted packets, and overflow packets. When the game play starts to suck, you can pop this up and get an idea of what the cause is.
Finally, if there are any FPS gods in the SF Bay Area who would like to help me graduate from frustratingly mediocre player to good-enough-not-to-embarrass-myself player, I'd appreciate hearing from you.
Schwab
Editor, A1-AAA AmeriCaptions
== Paul Rickard, Editor of The Microsoft Boycott Campaign ====
PHB's would love you to think that the internet is used primarily for business email (upper management email tends to be MORE important btw. even if it is to schedule the wife's baby shower) and for people surfing corporate sites for *valuable* marketing information.
As if that requires any real bandwidth or requires low latency/packet loss/jitter.
99% of the Internet is porn and games. Porn doesn't require low latency, low jitter or low packet loss, and can be safely QoS'ed into the "available bandwith" slot along with Mr. CEO's VITALLY IMPORTANT email to his golfing pals.
The only thing left is games and VoIP.
The latter is strictly CBR.
This leaves GAMES.
Nobody wants to admit this. It is the Internet's dirty little secret that when a company complains to an ISP about its shitty latency and packet loss rates, it is NOT because Mr. PHB can't check his stock portfolio (after all, he can do this over a modem with 20%+ packet loss and a ping of 500 ms). It is because somebody in the IT staff just got fragged by an LPB.
I don't follow how he got the measure of 150ms as what people prefer. What the data really shows is that most people who connected to his quake server had a ping time of under 150ms (which actually covers most of the US from the plots he drew).
Let's think about this a little more:
i) He's on a T1 and possibly advertises the fact.
ii) Serious MP Quake players will have a fast connection (DSL/Cable) and not a modem.
iii) Quake players will *choose* the lowest ping because it seems like a good idea - they won't reject a server over 150ms, just prefer one below it.
Looking at the results, the best he can say is that most people that play on his server are from the US, which for the most part keeps their ping under 150ms. You can't draw any assumptions about what people "prefer" from that data.
Fear: When you see B8 00 4C CD 21 and know what it means
Based on analysis of a QuakeIII server over two months in early 2001, first person shooter (FPS) game players strongly prefer servers that are within 150 milliseconds of the player's attachment point to the Internet. Or put another way, if you are planning on hosting a for-fee FPS server service, you should forget about trying to get customers who are more than 150 milliseconds away from your servers. Sections of the Internet more than 150 milliseconds away will provide a minimal source of regular players (and hence minimal revenue).
I fail to see how this indicates some kind of acceptable limit. A similar study done 2 years ago might have reached exactly the same kind of conclusion for 200 msec. Within a year, it might yield 100 msec.
All this tells us is that most players succeed at getting 150msec pings if they can. It doesn't tell us whether they find this number satisfying. If they were unable to play at 150, they might very well be happy playing at 200 or 250.
So this doesn't tell a potential service provider anything beyond the current status of the internet's average ping times. And such data will be useless in a year.
What would demonstrate his point would be players who didn't know what their ping was, and were able to switch between servers with known pings in a controlled situation (50/100/150/200... for example). One could then watch where the players end up, if you saw no difference between the 50 and 100 ping servers, you could *then* conclude that players were satisfied with 100 and saw no need to try to get to 50.
The thing is... A game with robustly coded prediction (such as Q3) can do reasonably well with times up to 300ms. The problem is that as your ping gets above 150ms, you get a lot more packet loss as well. And when your packet loss spikes, prediction is MUCH harder. If I were paying for a gaming oriented ISP, I would put guaranteed throughput above low ping on my priority list.
Of course, games whose prediction engines are notably worse (*ahem*UnrealTournament*ahem*) will suffer both from packet loss and high ping.
-Ted
> Thank you :)
:)
;-)
t ure04/Slide22.html
u rces/Shading/21.html
n gApplet.htm
You're welcome
> All these things I had been pondering before, but it's very hard to do a web search on "color depth" or "frame rates" and get useful results
Aye, you won't find the answers unless you knew what you were looking for, but if you knew what you were looking for, you wouldn't need to look. Or something like that
You can see an example of "Mach Banding" here
http://graphics.lcs.mit.edu/classes/6.837/F00/Lec
This page shows how our eye percieves Mach Banding
http://www.loria.fr/~holzschu/cours/HTML/ICG/Reso
And this applet lets you try it out:
http://www.cs.umb.edu/~ram/ArtAndVision/MachBandi
Cheers
> The human eye can only recognize a color depth of X, where X is less than current standard "true-color" depths. Yet we still have 32 bit color. Why?
:-)
Short anwer:
Where's the COLOR faq when you need it?
Seriously,
Gamma ties into this, which I'll ignore, since it a different problem, but the short answer is: you want as many colors as possible when you need to interpolate between any given 2. e.g. a high gradient.
At 32 bit color, with 8 bits per channel (e.g. 3 channels = red, green, blue) that gives us 256 levels for primary gradients. (Less for non-primary colors.)
32 bit is still too low, though. Ideally we should have 16-bit per channel (i.e. 48 bit color) since our eye is more sensitive to greens, and 256 shades isn't quite enough.
So it's not about TOTAL colors, but about the NUMBER of color steps BETWEEN colors.
Blending, or "overlaying" transparent textures is another reason high bit depths matter. If you overlay 8 transparent objects in 16-bit color, you get [bad] artifacts. You can see this on the older Voodoo cards, when you had transparent smoke.
And finally, memory access if MUCH faster if memory is aligned on a power of 2. 24 is not a power of 2, while 32 is. Memory controllers are slower if they have to access memory on odd alignment. There was a paper few years ago paper showing you a "wierd" memory touching walk where instead of doing
for (int i = 0; i < 100000; i++ )
block[i] = c;
it would be faster to do
block[i+0] = c;
block[i+2] = c;
block[i+1] = c;
block[i+3] = c;
> will having our color depth set to 32bpp instead of 24 make a difference,
Aside from the major speed difference, visually no, because the last 8 bits in 32bit color are usually used for alpha (the level of opacity)
> So the question is, even though you can only see one third of those 210 frames per second, does that make your playing more enjoyable or better?
It's an expontential curve of decreasing returns.
i.e. double the frame rate from 10 to 20, is MUCH more noticable then the double jump from 30 to 60.
But there are a few reasons you want > 100 hz frame rates.
a) The more people that are on screen, the lower the frame-rate. You want a high frame rate so when the action gets "thick and heavy" your frame rate still is above 60.
b) I find a monitor with 100 hz to be rock solid and easy on the eyes. At lower frequence (like 60) Hz I get a head ache (probably because the way our body clock is tied to 60 hz)
> Similarly, does looking at a 48bpp image make you happier than looking at a 24bpp image,
You get less "Mach banding" with a higher bit depth.
Again, this is another example expontential curve of decreasing returns. 48 bpp is "very good". 24 bpp is "good enough"
Hope this helps.
Now we can properly categorize people regardless of their ISP or method of connecting to the Internet. We have a number. If you are less than 150, you're an LPB. Otherwise, you're just a p1ngk1dd13.
It's kinda like penis length, really. It doesn't matter at all in the grand scheme of things, but it's nice to have a number to compare yourself to.
Got Rhinos?
Ping times no longer measure network speed except in very highly congested networks. For the past 5 years networks have been fast enough that the largest contributor to round-trip times is simply the speed of light. (in a fiber-optic cable, or the speed of an electronic signal down a copper wire, either way about 2/3 c).
High ping times making games unplayable mean that north americans will never enjoy playing against australians, but it really has very little to do with the quality or cost of their network connections.
Tarsnap: Online backups for the truly paranoid
Not always the case. I've switched ISPs twice as a result of high latency. My beef is that using remote editors through Telnet and SSH is virtually impossible when it takes a second or more for each keystroke to show up on the screen.
Now did I type seven backspaces or eight? The cursor has stopped moving; is that all, or are there more cursor commands in the pipe?
--
Slashdot monitor for your Mozilla sidebar or Active Desktop.
I figure a lot of it has to do with a) what we are used to. And b) how fast we are as people.
:)
If you play at a lan party all weekend, and go home to your 80 ping, you swear you're playing in mud. Your rails are off, everything feels Wonky. And yet you may, if you're like me, remember a time where you were Completely comfortable living inside of a 250 ping. 'Getting inside of your lag' we'd say. And after 6 or 8 hours playing quake at 250 ping, eveything in real life seemed a little off, like it would be ok if you could adjust the latency a little
The other thing is, some people are just slower than others. If you're kinda slow, then you might not notice the difference between 150 and 200ms, simply becasue your body doesn't work like that. On the other hand, i know some squirril-people that may be able to detect network latency differences within a few ms no problem.
So i dunno. I guess the article is talking about a law of averages, but i think there is a lot more to consider.
Jacobson, V. 1990a. "Compressing TCP/IP Headers for Low-Speed Serial Links," RFC 1144, 43 Pages (Feb.). Describes CSLIP, a version of SLIP with the TCP and IP headers compressed.
No matter what my ping time is, I still keep blowing my stupid self up with the rocket launcher. :(
The more interesting time is to correlate ping times to suckage. I remember the Quake1 days where LPB's were very highly regarded -- and whenever one appeared on the server, all the modem users would hit their little chat keys and gripe about the LPB that just joined the game.
"I'm on a T1, guys," the LBP would say and would -- under most circumstances -- dominate the game. There were (at least from my vantage point as a modem gamer playing with mostly other modem players) very few LPBs, and those that I recall playing with -- especially in the Q1 days -- were pretty good and knew how to exploit their LPBness.
Now, however, it's a whole different story. You got SDSL users, wacky @home kiddies, whiny ADSL users, and the odd Starband user who invariably lands in your game, starts chatting about pings ("Hey, guys, I'm on Starband! Look at my terrible ping!") and then starts wanting to know you can see his new skin ("Hey, Kelso, can you see my new Skin?").
Ping times have gone down, but suckage -- at least from where I sit on my SDSL -- has gone way, way up. Which leads me to believe that the overall on-line gaming ain't what it used to be. When suckage is so high and pings are so low, you get discouraged.
Campers never bothered me, BFGers never bothered me, but there's nothing worse than a LPB that sucks.