pfarrell;346592 Wrote: > Have you measured it with collisions? Back in ancient times, there were > huge technical debates over token ring vs Ethernet. Because token > ring's performance with contention was essentially fixed, and > Ethernet's is probabilistic.
Yeah but ethernet won didn't it?! The reason is that although it's internals aren't as nicely & cleanly organized as token-ring, it slapped around the tokens' ears with better performance. Typical new bad kid in town that wins hands-down in the end because of result. Sure I measured it with collisions (you can't have it any other way with wifi!) and apart from some wifi-specific-encapsulation, it performs exactly like wired ethernet on a point-to-point link. CSMA/CD is a mechanism that deals with collisions and successfully avoids them for most of a point-to-point link. But there's more: > CSMA/CD never delivers more than half its rated capacity in any real > world case. a 25% delivery is good, and one tenth is not at all > unusual > with collisions being detected. Now we get to the half-duplex vs full-duplex stuff, and that's where the 50% comes from. In a half-duplex setup you only get 50% when the other 50% is used for return traffic and indeed you will get less than 50% because of collisions! I typically had 100BaseTx full duplex connections do 200Mbps continuesly 24 hours a day on our ISP usenet servers. So, full duplex doubles the total bandwidth but simplex doesn't halve it per se... it only does that when you have the return-trafic which you don't have on a typical "1-hop" audio feed. The only return-traffic is the TCP/IP ack/nack error-recovery mechanism which is basically nothing compared to the audio-stream. Collisions eating away bandwidth only happens when you have many speakers on the channel/wire like in a big office-building with hundreds of PC's all doing heavy networking with a file-server while listening Internet radio etc. That's where the ethernet switch comes in and now every PC (or small group of PC's) have their own wire. In a point-to-point half-duplex config you have little collisions as the CSMA/CD takes good enough care of it. So, on our typical home-network 54g wifi kit that's installed good so that all devices have good signal/quality levels and do 54 Mbps mode, you will have about 50 Mbps to share. That doesn't get halved because it isn't full duplex, you just share it. So if SC uses 2Mbps and 2 SBs together use 2Mbps you loose 4 Mbps and have 46 left. The numbers I state here are good rules of thumb for flac-encoded redbook and so include TCP/IP and wifi overhead. Add a couple of computers like son doing his bit-torrents and all-out-on-line-shooting game, it will all still work when you add yet another couple of SB's. But by that time it would be better to run a 100BaseTx from SC to the router. The LAN ports on the routers are switched BTW and many will do full-duplex. My DLink4500 has 4 gigabit full duplex switched ports which can do a total throughput of 8Gbps (4 up + 4 down) which it's silly silicon parts won't handle ;-) If you get so much stuff that you get wifi congestion, you just buy some 5Ghz n-adapters for the pc's and you open up a 2nd frequency-channel with even more bandwidth than the first and the out of the box DLink4500 will do it. When you enable any encryption like WEP or WMA you --will-- loose big time, not because of the wifi limits but because the routers cpu can't handle it. Also, devices like SB tend to have trouble with it as can be read in the forums. Use MAC-address access-lists in the router instead to keep strangers off your network. Your banking etc. stuff is still encrypted when the URL says https:// but in a big-city neighbourhood I would either run a cable to the router for that kind of stuff or put up a second wifi router with encryption just for that... it's cheap enough! BTW, I do remember that a news-server with half duplex 10BaseT (10 Mbps) connection to a switch-port could do 8 Mbps one-way feed so that was a loss of 20% due to sharing the link (TCP ack/nack) and collisions. TCP encapsulation is dependant on the size of the TCP packets. The headers are fixed so if you pack more bits in each packet you win. Most heavy-bandwidth stuff incl. audio-streaming will go for big packets so count a 10% overhead there. 1 Mbps for a flac encoded redbook is conservative as it'll use less but a nice and safe number to work with. Testing is easy, just copy a big file between a wired and a wifi pc/nas and time it. For fun, take a 1-minute flac of redbook audio and see how many seconds the copy takes. Or use calculator for a 3.47 minute track. I'll test like that and report in this thread. Also, I want to mention another often made mistake: people tend to max-out the wifi power-settings on all devices. That's counter-productive because the singal from client-device A will add noise to client-device B's link. Reduce power where-ever possible. The best way to reduce noise is to put the wifi-router not in the center of the house but on 1 side, towards where most neighbours wifi-routers are. Now add a 180-deg directional external antenna pointing through your house. This cuts out all (well, no antenna is perfect so not all but it will dampen these many dB's) neighbours signals to your router or at least half of it (the other 180 deg.) Most routers take external antenna's and they are cheap. It's like audio: singal-to-noise ratio is what counts for good quality link and cutting noise is better than increasing signal! cheers, Nick. -- DeVerm ------------------------------------------------------------------------ DeVerm's Profile: http://forums.slimdevices.com/member.php?userid=18104 View this thread: http://forums.slimdevices.com/showthread.php?t=53018 _______________________________________________ audiophiles mailing list [email protected] http://lists.slimdevices.com/lists/listinfo/audiophiles
