I never said 27 Mbps was reserved, but in most normally observed scenarios it seems to be a roughly 50/50 split

Gives up on Wifi

In reply to a post by MrSaffron:

I never said 27 Mbps was reserved, but in most normally observed scenarios it seems to be a roughly 50/50 split

Gives up on Wifi

Its ok, don't worry, I'm obviously failing to get it across, it seems clearer in my head (The point is there is no splitting of the link rate! One direction has one link rate - eg 300Mbps, the other has its own - could also be 300Mbps, could be 60Mbps . It's a common normally observed scenario for them to be significantly different eg at range, eg with 1x2:2 clients)

In reply to a post by vimto_girl:

Its ok, don't worry, I'm obviously failing to get it across, it seems clearer in my head (The point is there is no splitting of the link rate! One direction has one link rate - eg 300Mbps, the other has its own - could also be 300Mbps, could be 60Mbps . It's a common normally observed scenario for them to be significantly different eg at range, eg with 1x2:2 clients)

This is fascinating - I'm wondering if you're both approaching the question at a different angle. One from the TCP/IP throughput, and one from the radio Tx/Rx side.

My own testing on various laptops (1 stream, 2 stream, 3 stream) with a three stream router has shown that G throughput is along the lines Mr Saffron mentioned. This may be due to TCP/IP getting delayed ACK packets due to RF interference - but 802.11g is never capable of 54 Mbps real world TCP/IP throughput. Similarly HomePlug AV200 or AV500 adaptors don't reach that speed at the TCP/IP traffic level. Only copper Ethernet gets the speeds described.

edit - spelling

Edited by jchamier (Sat 22-Feb-14 18:39:41)

Thank you.

And am aware that if a connection has a download link speed of 80 Mbps reported by WiFi card, that is not split in half. The reality though for Windows users is usually the combined figure of download+upload link speed is reported, where its shown in OSX is something I need to figure out

In reply to a post by MrSaffron:

where its shown in OSX is something I need to figure out

Hold Option (Alt) key and click the wifi icon in the menu bar. You'll get "Transmit Rate" which on my 2012 Air shows as 300. (two streams 40mhz 5GHz N as per wifi certification, on my 3 stream capable Asus).

(don't ask me about Linux, each distro will be different, and all my Linux boxen are server hardware, or VMs.)

(I'm only replying cos you said fascinating - if anyone is not interested, they don't have to engage in the conversation anymore). So....I don't think what you say is the source of the difference - we both agree that a typical max TCP throughput in any one direction for say 802.11g is 22Mbps and that the max link rate is 54Mbps. The radio chains are just an example to illustrate my point - I can make my point with single-radio 802.11g just as well - so this doesn't explain it.

My best interpretation of MrSaffron so far is that say you have a 802.11g client that declares on the box 54Mbps. 54Mbps refers to the maximum link rate, and when achieved, this means the transmit link rate plus the receive link rate must equal 54Mbps. Because this is usually a 50/50 split, the transmit link rate is 27Mbps and the receive link rate is 27Mbps - and this accounts for the bulk of the throughput 'deficit'.

I'm saying the 54Mbps refers to the maximum link rate. At any point in time a client has a Tx link rate and a Rx link rate. The two link rates can both be 54Mbps, one can be 54Mbps and another 12 Mbps, and so on. The throughput deficit is not because the 54Mbps link rate must be shared between both directions, but due to overheads - minimum transaction time (acknowledgement even in ideal RF environment) being the bulk when we are not sharing 'airtime', plus all the others. If we are just sending traffic in one direction, the link rate of the other direction does not affect it - it could be many times more or many times less or exactly the same.

Talking about a 300Mbps client "NO The 300 Meg, is actually 150 Meg in one direction PLUS 150 Meg in the other, and then the overheads need to be taken into account". I said this isn't right, in quite a fundamental way - not nitpicking, and a 1x2:2 300Mbps client is a vivid example that exposes this.

You're absolutely correct. WiFi is half duplex w/ TDD.

I imagine if you used a protocol like UDP over WiFi you'd see much better results. You wouldn't require ACKs going the other way as in TCP and that will reduce the amount of time that the Tx has to be using airtime up for.

Ta - used OSX for year and a bit but still miss the hover hints of windows

In reply to a post by vimto_girl:

I'm saying the 54Mbps refers to the maximum link rate. At any point in time a client has a Tx link rate and a Rx link rate. The two link rates can both be 54Mbps, one can be 54Mbps and another 12 Mbps, and so on. The throughput deficit is not because the 54Mbps link rate must be shared between both directions, but due to overheads - minimum transaction time (acknowledgement even in ideal RF environment) being the bulk when we are not sharing 'airtime', plus all the others. If we are just sending traffic in one direction, the link rate of the other direction does not affect it - it could be many times more or many times less or exactly the same.

Thanks, that's a great explanation (as is mr_mojo's mention of half duplex).

The problem is the user interface on most software only shows a single "link speed" entry, as it was designed for copper Ethernet. Back in the day we used 10 Mbps half duplex (and even 100 Mbps half duplex) the connection always showed 10 or 100 on this indicator.

Its people's expectations that a 54 Mbps status indicator for WiFi means its the same Tx and Rx, which in WiFi it has good reason that might not not be.

thanks.

In reply to a post by MrSaffron:

Ta - used OSX for year and a bit but still miss the hover hints of windows

Its a nice OS, but often things that are easy in Windows are either easy in OS X or impossible. Sometimes its worth getting low level and grepping logs (as you do on Linux systems), such as when my MBA decided my router wasn't UK band compliant. (luckily a firmware update fixed that).