Our test uses large packets and TCP. The problem is that the bandwidth tester consumes a lot of the CPU, so the solution is to test between machines on the edges, and thus get the true throughput without the limitations of CPU speed.
Lots of guys have used desktop machines and found that the actually Atheros output can exceed 100 mbps in Turbo mode. Of course that sort of test cannot be done through an Ethernet, it is radio to radio. The advanced features of the Atheros cards will do packet aggregation for the system. Thus you will not notice any system trouble with a bunch of small packets as you get with VOIP and gaming. That type of traffic can bring a Prism system to its knees. The new StarVx also honours the VOIP priority bits in the headers. Lonnie On 10/10/05, David E. Smith <[EMAIL PROTECTED]> wrote: > > > I just installed a link using CM9, Station Server, WRAP board, about > > 300 yards away LOS, with only a single client so far on AP to SU > > mode. The radios associated at 54 mbps, and about -70 db, with a > > quality of 24/29. All speed enhancement features enable, and > > encryption turned off. > > > > Using Station server throughput test, testing from AP, the RX was 13.6 > > mbps, and the TX was 9.1 mbps. > > Sounds about right, give or take a couple Mbps. I've got one link > consisting of two WRAP boards, StarOS/WRAP edition, with CM-9 radio cards. > Signal is about -60, noise floats around the mid-90s. The two endpoints > are probably about 1/4 mile apart. > > StarOS's integrated bandwidth test shows about 14Mbps each way. (You have > to do both send and receive tests, from both units, to get good average > numbers, because sending the packets seems to seriously stress those > little tiny CPUs.) > > In my experience, it seems as though StarOS' numbers are very optimistic > (anywhere from 10% to 25% higher than real-world traffic). Having never > sniffed the traffic or anything, I'd guess that the bandwidth test uses > large packets (maybe even jumbo packets), but real-world IP traffic has a > lot of smaller packets, and the per-packet overhead brings down total > performance. (This is part of why many people are fond of Mikrotik's > proprietary Nstreme extension - if you can live with a couple extra > milliseconds of latency, you can get substantially better throughput by > bundling all those tiny VOIP and UDP packets together.) > > When using the StarOS built-in tester, the sending unit's CPU will hit > 100%, and the receiver's CPU hits 75% or higher, on WRAP boards. This > implies that the CPU may be the bottleneck, not the radio card or the OS. > > > My tech question is... Is this being limited by the Atheros chipset, > > or the WRAP motherboard? If using the Mikrotik RB532 board with higher > > processing speed, can a single Atheros card transfer at a higher rate? > > Get back to me in a week or two on that. :) I've gotten my hands on a > couple RB532s, and a couple extra CM-9s. Admittedly, it won't be a > completely fair test, since the RouterBoards will be running RouterOS, so > it's not a perfect comparison to Valemount's StarOS. But it's the best I > can do. > > Since the WRAPs don't have that much horsepower to begin with, in relative > terms, and since these tests pegs the CPU gauge, I suspect CPU is the > limiting factor, but I don't have any hard numbers to back this up either > way. > > Anyone picked up one of Valemount's new WAR boards and their modified > StarOS distribution for them (StarVX)? For that matter, anyone ever just > slapped a miniPCI-PCI converter in some cheap Dell desktop and slammed > packets through it? > > David Smith > MVN.net > -- > WISPA Wireless List: wireless@wispa.org > > Subscribe/Unsubscribe: > http://lists.wispa.org/mailman/listinfo/wireless > > Archives: http://lists.wispa.org/pipermail/wireless/ > -- Lonnie Nunweiler Valemount Networks Corporation http://www.star-os.com/ -- WISPA Wireless List: wireless@wispa.org Subscribe/Unsubscribe: http://lists.wispa.org/mailman/listinfo/wireless Archives: http://lists.wispa.org/pipermail/wireless/
