802.11a - 54Mbs @ 5Ghz?

[ginahoy]

As a new member, I wanted to say how much I appreciate this resource. Although I have a technical background, I haven't paid much attention to wireless networking developments. I'm pretty much a novice. So here's a novice question:

My b/g router just died. I'm using DSL modem as a router temporarily, but it's pretty awful. So I've been looking at N-class routers. I live in a fairly dense neighborhood with a dozen or so WLAN's and who-knows-how-many 2.4Ghz phones, baby monitors, etc, so I'd like to switch to 5Ghz band. However, I've only been able to find a couple of N-class cards that will fit my laptop (mini-pci), and neither supports 5Ghz.

With my b/g router (set to g only), I was able to drag-and-drop files on my LAN with sustained throughput of about 18Mbs, which I understand is typical. However, the best internet download speeds were around 2Mbs, compared to 5Mbs via Ethernet (limited by ISP). This suggests wireless imposes a 60% overhead hit. So I'm not convinced upgrading to N would provide much benefit on internet downloads. Presumably N has similar overhead as G... I guess that's a question.

I noticed that 802.11a operates at 5Ghz, and has max speed of 54Mbs. Given the amount of traffic on 2.4Ghz, I'm wondering if I wouldn't be better off with a legacy A-class router operating on 5Ghz than a N-class operating on 2.4Ghz. Range is not an issue in my situation. I verified my a/b/g NIC supports 5Ghz (there's a checkbox on the label).

What say you?

 

[ginahoy]

Anybody? You folks are the experts!

 

[stevech]

If your WiFi connection speed reported is very much higher than your ISP's speed, and if your throughput test method to a distant host is valid, then WiFi should not be the limitation.

With 11g and an ideal signal and no sever interference you should get 20-25Mbps net yield at the IP layer.

Using speedtest.net to remote hosts is, of course, statistic on which host you choose, time of day, etc. So choose different hosts and take averages.

Better to assess WiFi by doing LAN transfers among PCs. This takes the variability of the Internet out of the equation.

I would not go to 5.8GHz unless you are sure that you have incurable interference among the 3 viable channels in 2.4GHz (1, 6 11)

 

[ginahoy]

Thanks for your reply. This issue has been bugging me for a long time. For reference, my laptop is a T42 Thinkpad and my router is (was) a Linksys WTR54GL. Here are some throughput examples:

TEST....................802.11g....Ethernet
DSLR flash speedtest......250.........675 (KB/s)
filefront download........260.........710
local PC-to-PC...........2225........9900

These results are typical, based on a more than a dozen tests with the WTR54GL. I didn't see that much variance for this particular set of tests. For example, the internet download via ethernet might drop to 650, but the wireless download also dropped proportionally to about 240. I always ran these tests in sequence at different times of day. I've also tried different channels, and tried with security disabled. I even replaced my wireless card (2200bg). None of this made more than a few percentage points difference. Since my Linksys died, I'm getting similar results with my DSL modem/router (Actiontec PK5000).

What these results suggest is that the G protocol takes more than 50% off the top. I assume everyone else sees this kind of hit? In the case of PC-to-PC transfers, the ratio is even lower but in this case it's because the WLAN hits its limit, which as you point out, is more than 50% lower than its 54Mb physical capacity.

If my overhead hypothesis is correct, then the question becomes whether an N-class WLAN imposes a similar overhead burden.

You see what I'm getting at?

BTW, I've all but discarded the idea of 802.11a. Apparently not many consumer routers supported 11a, and the ones I identified got bad reviews (e.g., D-Link DI-784, Belkin F6D3230-4). Also, my PC-to-PC results suggest interference is not an issue, so my desire to move to 5Ghz was misguided.

 

[thiggins]

See SmallNetBuilder's Wireless FAQ: The Essentials
FAQ 1 for a table of typical wireless speeds.

Yes, wireless has a large overhead vs. wired, no matter what the protocol. With N, you just have higher maximum link rates, so the end-user throughput also is higher.

 

[stevech]

WiFi net yield at the IP layer is typically 60-70% of the reported 802.11 link speed.

The tests results you show- are suspect. What test tool did this? What produced it? It says your DSL is 6xxKBps - the "B" is supposed to mean Bytes/second, but in that context, it must be bits (little b) per second. Are you paying for 1.5Kbps (down) DSL?

The WRT54GL, old though it be, should be fine for DSL speeds.

 

[ginahoy]

wireless has a large overhead vs. wired, no matter what the protocol.

Thanks for your reply, Tim. Since maximum internet speed is just the starting point for the WAN-WLAN conversion, it seems like any difference in overhead would be the determining factor for maximum throughput for internet downloads. That's what I'm trying to investigate.

In the table below, the second column is theoretical physical layer from 802.11 specs. The third column is best-case empirical data from table at FAQ 1:

1 stream
802.11b/20Mhz.....11Mbs.......4-6Mbs....36%-55%
802.11g/20Mhz.....54Mbs.....20-23Mbs....37%-43%
802.11n/20Mhz.....72Mbs........40Mbs....56%
802.11n/40Mhz....150Mbs........60Mbs....40%

2 streams
superG/40Mhz.....108Mbs.....35-40Mbs....32%-37%
802.11n/20Mhz....144Mbs........70Mbs....49%
802.11n/40Mhz....300Mbs....80-100Mbs....27%-33%

BTW, I just conducted several download tests in 11b mode. As expected, results were very similar to 11g - about 36% throughput compared to downloads via ethernet. That's almost 2/3rds of my internet capacity lost to wireless overhead!

Interestingly, the table suggests overhead is lowest for 11n in 20Mhz single-stream mode, with hypothetical throughput of over 50%. For LAN-WLAN transfers, 40Mhz would be faster, but if my hypothesis is correct, anyone concerned with internet download speed should try using 20Mhz mode.

Admittedly, my analysis is a gross oversimplification of what's actually going on. But what seems to be missing from the discussion of wireless performance is empirical data on internet download throughput as a fraction of internet speed. I understand stevech's point that this introduces variability, but if download speed is what you're interested in, then that's what you need to measure.

 

[ginahoy]

The tests results you show- are suspect. What test tool did this? What produced it?

The first test is here. The second test is a large file download from filefront. I watch browser's progress display (Firefox) and I wait about a half minute to allow speed to stabilize. The third test is drag-and-drop in Windows Explorer (XP SP2). Since Explorer's progress display doesn't show transfer speed, I time the transfer. Finally, I sometimes use DU Meter so I can compare number of bytes transferred (both directions) with file size, just to confirm I'm not losing a large number of packets. I typically see less than 3% difference.

It says your DSL is 6xxKBps - the "B" is supposed to mean Bytes/second, but in that context, it must be bits (little b) per second. Are you paying for 1.5Kbps (down) DSL?

No, when I write capital B, I do mean bytes per second. For example, in the above table, 675 KBs is the same as 5.4Mbs, typical for my DSL connection. I pay for 7Mbs (bits) but the highest I ever see is about 5.9Mbs, or 735KBs.

The WRT54GL, old though it be, should be fine for DSL speeds.

As I said, the WRT54GL died so I'm trying to decide whether to buy an n-class router. In particular, I don't want to go through the hassle/cost of replacing the NIC unless there's a payback. The laptop belongs to my wife. She recently started downloading large training videos related to her profession. Also, we've been talking about buying a streaming Blu-Ray player, which unfortunately would have to be wireless.