Re: Throughput less than expected: madwifi 0.9.4 r4119
"Brian Prodoehl" <[email protected]> Mon, 5 Apr 2010 08:09:18 -0400
| Newsgroups | gmane.linux.drivers.madwifi.user |
|---|---|
| Message-ID | <[email protected]> |
I think your math is still off a bit, but that's a great start. Quite a bit of the overhead is sent at the base bitrate, and ACKs are sent at a modulation independent of the original packet. I can't imagine that increasing the size of the transmit queue will do much more than increase the delay between when the packet stream starts and when you start dropping packets. I would suggest experimenting with the contention window parameters (cwmin and cwmax), and also the ACK bitrate (ackrate parameter). I have experimented with both, with varying degrees of success, but don't remember enough off the top of my head to suggest anything specific. The MadWifi wiki, and google, is where I would start. Dropping cwmin and cwmax, and increasing the ack bitrate, should help increase your packet rate. Brian Prodoehl -----Original Message----- From: R.Mohan [mailto:[email protected]] Sent: Sat 4/3/2010 3:02 PM To: Brian Prodoehl; [email protected] Subject: RE: [Madwifi-users] Throughput less than expected: madwifi 0.9.4 r4119 Dear Brian, I was away for a while and could not reply to your mail earlier. My earlier math was erroneous. I concur that the max frame rate can be a shade higher than 6K frames per second. I have attached my revised calulations to this mail. I am assuming that I will be using the Atheros burst mode as well as ACK for every frame sent. I am still getting only approx 2K frames per second (frame size does not seem to matter). At higher rates athstats reports that tx frames are discarded due to Q depth. I still need help in increasing the Q depth or any other fix so that I am able to substantially increase my frame rate. R.Mohan -----Original Message----- From: Brian Prodoehl [mailto:[email protected]] Sent: Wednesday, March 24, 2010 12:18 AM To: [email protected]; [email protected] Subject: RE: [Madwifi-users] Throughput less than expected: madwifi 0.9.4 r4119 Can you share the math that leads you to believe that 8,000 packets is possible? 802.11 carries quite a bit of overhead. This may help: http://www.oreillynet.com/pub/a/wireless/2003/08/08/wireless_throughput.html I don't know how even your packet size distribution is, but plugging in different numbers, I come up with something like 6,000/s being theoretical max for you, and that's before needing to deal with contention for the medium. I would guess that real-world over-the-air your limit would be something like 3,000/s without some drastic changes to the MAC and also tweaks to the PHY parameters. Brian Prodoehl -----Original Message----- From: R.Mohan [mailto:[email protected]] Sent: Tue 3/23/2010 1:22 PM To: [email protected] Subject: [Madwifi-users] Throughput less than expected: madwifi 0.9.4 r4119 Dear All, I am experiencing only 25% 0f the expected throughput. Athstats reports "tx frames discarded due to queue depth" Can anyone help me to increase the queue sizes / resolve this problem. I would like transmit 8000K plus frames per second. Frame sizes will range from 78 bytes to 558 bytes. Presently I am able to transmit approximately 2000 frames per second without any tx frames being discarded. Below is the report from athstats: # ./athstats 84 tx management frames 1952979 tx frames discarded due to queue depth 1418 short on-chip tx retries 978 long on-chip tx retries 70 tx frames with no ack marked 303132 tx frames with rts enabled 303146 tx frames with short preamble 109 rx failed due to bad CRC 606261 rx failed due to frame too short 2347 PHY errors 142 OFDM timing 1 OFDM illegal rate 1790 CCK timing 21 CCK header crc 326 CCK illegal rate 67 CCK restart 9 periodic calibrations rssi of last ack: 51 rssi of last rcv: 45 1 switched default/rx antenna Thanks in advance R.Mohan ---------------------------------------------------------------------------- -- Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Madwifi-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/madwifi-users ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Madwifi-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/madwifi-users