Performance numbers [Archive] - Valemount Networks Support Forums

tony

10-28-2003, 08:50 AM

Our new Atheros support is progressing nicely, and are interested in what others are seeing as far as throughput on existing Atheros-based systems. (such as 802.11a/g APs or other software products, etc.).

The numbers we are seeing with our new driver are quite good, however we do not have any basis for comparison.

We are interested in these numbers:
@ 802.11a (AP) <-> 802.11a (client) throughput tests in KBytes/sec
@ 802.11a Turbo (AP) <-> 802.11a Turbo (client) throughput tests in KBytes/sec
@ 802.11g (AP) <-> 802.11g (client) throughput tests in KBytes/sec
@ 802.11g "108Mbps" (AP) <-> 802.11g (client) throughput tests in KBytes/sec

Please mention if these tests are in a closed test environment, or actual deployments.

If you have numbers for non-atheros 802.11a and 802.11g products, those numbers are also welcome.

Thanks!

mp3turbo

11-05-2003, 02:23 AM

Hello Tony,

I'll give you those numbers tomorrow - I don't have AP and antennas yet, so we were trying indoor ad-hoc only on AB520 Dlink PCI adapters. Basically, we got more than 2.5MB/s TCP/IP throughput using FTP (client-server) and Windows File Sharing. This is the same we got using 802.11g 2.4GHz adapters from SMC and other Taiwanese manufacturer.

Do you plan to have Super A (or 802.11a Turbo 108Mbps) in your drivers? Would it be possible to use outdoors?

bye, Lubomir.

tony

11-05-2003, 07:52 AM

There will be support for the 'A' and 'G' 108Mbps modes in the initial release.

Thanks!

mp3turbo

11-12-2003, 12:21 AM

hello tony,

do you have the specification and maybe some spectrum analyzer to confirm that 802.11a turbo mode takes 2x 20mhz bandwidth? Isn't it more?? Based on my measurements with enterprise-grade hardware (but not spectrum analyzer, it was just enterprise class radio system), standard bandwidth used by 802.11a normal mode is MUCH wider than stated 20MHz (I was able to find 40MHz !! of signal spread with pretty intensive power!!) - anything to this?

thnx, lubomir.

bobbyc

11-12-2003, 12:51 AM

Probably; there's most likely some sidelobes extending out from the 20mhz.
108mbps (turbo) will be nice for interoffice type of networks, but for our outdoor stuff I think I am going to shoot for 24mbps (-75) or 36mbps (-73).
We're looking at these cards:
http://www.fament.com/catalog/product_info.php/products_id/162?osCsid=44d7d96a73a49e2c3ce666c6817a268b

datawork

11-12-2003, 03:41 AM

szern

11-12-2003, 04:22 AM

11g: 8011 KBps :shock:
11g turbo 7996 KBps :wink:

802.11g @ 8011 KBps = KiloBytes per sec? so mutiplied by 8 = 64088 Kbps (Kilobits per sec)
Wouldn't that be in excess of 54Mbps ?

Just wondering...

- Szern

tony

11-12-2003, 08:03 AM

Thanks for posting your numbers.

These numbers seem accurate. The DWL-G520 (and all a/b/g cards) are Atheros 5212 based which provide performance enhancements that exceed 54Mbps actual throughput in many cases.

The Atheros cards are indeed nice, with great performance potential.

Thanks!

lonnie

11-12-2003, 09:25 AM

11g: 8011 KBps :shock:
11g turbo 7996 KBps :wink:

Both machines are ATX 733 with dwl-ag520 in AP Mode and dwl-g520 as client.

Tomorrow we'll give things a shot at 10 miles and see what we get.

Do you guys have anymore of these left ATX733 or are you moving towards the WRAP boards

Turbo doesn't do much if you have a lot of signal around. Since it moves to Channel 6 it is bothered by almost everything. If you could get a clean environment you would see 12,000 KBps, or simply run this on 802.11a Turbo and rock and roll.

Looking forward to the distance numbers. We have not yet tested externally, but that will happen on a couple of links next week. Time to stress test this driver.

We are caught short because the WRAP boards are taking longer than anticipated. The ATX sold out real quick because we kept the same price but still included Server License. The 10 pack basically gave the motherboard, ram and flash for about $20, so it was a no brainer, or as I like to call it - You Snooze You Lose.

bobbyc

11-12-2003, 10:39 AM

Wait, so you're saying that the .a numbers are much much much closer to the actual throughput.... IE 56mbps is near 56mbps actual throughput? Or is it like .b where 11mbps means 5mbps?
Bob C

tony

11-12-2003, 10:51 AM

Yes, .a and .g can reach, and quite often exceed 54Mbps of actual throughput using the ar5212 chipsets in non-turbo modes. The ar5211 chipsets (as found in the DWL-AB520) do not have these enhanced abilities, but still perform quite well.

Note that our tests are all localized and show best-case situations. Actual deployed systems will vary in speeds depending on environmental conditions. We are always interested in hearing what throughput people are seeing with their deployed systems.

bobbyc

11-12-2003, 11:01 AM

That's great news. We have the choice of doing a 32 or 40 mile link from our backbone to one of our hilltop towers... Whether we do 32 or 40 will depend on which hilltop has the least noise. I've done the SOM calcs and it looks like it could be done with 4' dishes. It would be nice to retire the bonded T-1s we have pulled up here now... or at least use 1 for backup to the wireless backhaul to our backbone.

Have you guys dove into ack timeout?
Bob C

tony

11-12-2003, 11:07 AM

The ack timeouts are being worked on and are quite complicated in calculation due to the varying speeds and modes available. Once we have this feature available, it will be tunable much like our Agere cards.

datawork

11-18-2003, 06:26 PM

Well, the results aren't pretty. Even with 1 watt amps at both ends and 24 db grids the highest rate we could pull of was 120 KBps. This same link with 100mw prism 2.5 cards averages between 300 and 400 KBps. 60' of lmr 400 at both ends which I think becomes a large factor with the 15db of output with the d-link cards. Anyone know if there are any 100mw atheros 11g cards being manufactured.

szern

11-18-2003, 06:39 PM

IIRC the Senao/Engenius 802.11g products (NL-3054PCI Aries & NL-3054MP Aries Mini-PCI) are rated at 100mW (20dBm)

- Szern

Pete_WirelessWales

11-19-2003, 08:19 AM

1 watt amps at both ends and 24 db grids

Are the amps able to cope with OFDM though? They may be specific to DSSS?

You shouldn't really need any amping with 24dBi at that distance - we have 802.11b links here running at 11mbps over a that sort of distance with a 19dBi dish one end, and what we believe is an omni the other!

lonnie

11-19-2003, 09:14 AM

None of the current 2.4 GHz amplifier are usable for 802.11g use. They get over driven and are very noisy. Try the link with no amplifier and if you get any sort of linbk I am sure you will get faster than the 120 KBps you are seeing with the 1W amplifiers. If the amplifiers are non AGC then you might have a chance by reducing the input way down until the output gets clean. If they are AGC then they will simply be over-driven.

10 miles with 24 dB antennas and 35 mW is no problem at all. The Atheros cards are typically approaching 100mW so I would really advise to remove the amps and try again.

datawork

11-19-2003, 12:42 PM

The amps were designed to be used with 2.4 OFDM and are 11b,11g compliant, they are not RFLinks.

The area we are in is very noisy and yes in theory the d-link cards with the grids should work but we had the same result with the prism 2.5 cards. No amps= poor throughput. We also tried this link initially with Orinoco cards and got nothing.

If anyone is interested in these outdoor amps, please let me know.
250, 500 mw, 1 and 2W versions are avaliable.

My appoligies for the shameless plug
bill@gi-wireless.net

datawork

11-29-2003, 02:18 AM

well we frinally recieved our mini-PCI 100mw Senao cards :lol: and slapped them into our D-link 520 pci boards. The results about a ten fold increase over the stock 520 cards but still throughput is only in the 180 KBps range with 11g 1 watt amps and 24 dbi grids. Also tested link without amps and the two cards wouldn't see each other. :cry:

Here are the specs from the association list on the AP side qual 44/42 Signal -52/-53 Noise -102/-102 Rate 2

Readings on the client side are Qual 39 Signal -56 dbm rate 2

60' of LMR 400, 24 dbi grids, 1 watt 11g amps 100mw 11g cards at each end.

Once again I must stress that we are in an extremely noisy area and these results are almost identical to what we saw with 100mw b cards. Things only improved (2-3mbps) once we upgraded our 11b cards to 200mw, I would waiger that the same will hold true if and when we ever see 200 mw 11g cards. Seano is rumoured to be working on such a beast.

lonnie

11-29-2003, 12:37 PM

Using 802.11g in a "busy" environment will not work too well. The chips have builtin logic to make them stay in 'b' mode if they detect much 'b' traffic so they do not interfere with the other 'b' units. They have to do this so they are not causing intentional interference to existing systems.

I think you might have to go to 'a' units. 10 miles with 21 dB panels will just about do, if you have the radio right at the antenna. Move to a 27 dB antenna and the numbers are good and at 31 dB antennas they are very good

datawork

11-30-2003, 01:32 AM

Lonnie,We are on the same page. I'm am waiting for a couple of the equinox 32 dbi grids, which are supposed to begin shipping Monday. We will be mounting them as low as possible on the towers (15') to cut down on wind load and line loss. Should give us a half decent fade margin.

In light of this we have a couple of the one watt 11g amps (not RFLinks or overdriven) with power injectors up for grabs at $900 CDN each, if anyone is interested. If not I'm sure we can put them to use elsewhere down the road.

lonnie

11-30-2003, 09:49 AM

Who has the Equinox stuff? I was told there would be nothing until the New Year.

datawork

11-30-2003, 10:14 AM

Just got a response from Equinox and they are going to try and ship them to us tomorrow.

bobbyc

11-30-2003, 01:30 PM

A 32dB grid antenna? On their website all they have is 26dB max for 2.4ghz and 28dB max for 5.8ghz. Is this new?
Bob C

bobbyc

11-30-2003, 02:49 PM

OK I found 32 and 36dB 5.8ghz models located on the ecommwireless site.
This is what I don't understand:
How can Equinox rate a 39" grid dish at 36dB gain, while a SOLID radiowaves 48" dish is only 34.6dB?
Are the Equinox grid dishes magic or something?
Bob C

datawork

12-01-2003, 12:55 AM

there is a misprint in the Elecro-comm PDF the 39" grid should be 32 dbi and the 32" should be 28 dbi. This has been pointed out to them.

datawork

12-03-2003, 12:38 AM

Talked Eqinox today and our 32 dbi antennas have shipped minus the feed horns that will be drop shipped on the 10th. You can order product from Electro-Comm and equinox will drop ship them now if requested when you place your order with EC.

bobbyc

12-08-2003, 12:33 PM

So when you are talking about shooting a distance shot (20-30) miles, I wonder if a solid 29dB pac wireless dish would outperform the equinox 32dB grid dish? I know that by looking at the antenna gain, one would assume that the 32dB grid dish would be better, but in my experience at 2.4ghz, the Andrew 18dB grid dish outperforms the pacwireless/equinox 19dB grid styles... the grids are thinner on the pac wireless dishes than the andrew dishes... and from the looks of the equinox 5.8 dish, its the same style of grid width.

Also a flat panel 19dB panel antenna outperforms a 19dB grid antenna at 2.4ghz.... further supporting that solid antennas are better?
Bob C

datawork

12-29-2003, 10:14 PM

Here are our initial throughput numbers and setup info.

Throughput: averaging 32 KBps :cry:

The interesting reading is the link quality which is around 18

signal -79
Rate 24

setup

site one AP

D-link AG520
15 ft LMR 400
29 dbi Pacwireless Para dish
Ack 500
Country CA
Rate auto ( have tried various hard coded rates but hasn't affected throughput)

site two client

Seano Aries AGB card
20 ft LMR 400
29 dbi Pacwireless Para dish

Any suggestions would be most welcome, from the link quality I am leaning towards a lack of power. I realize using LMR 400 isn't ideal with 5.7 but the lengths shouldn't be causing this kind of performance hit.

lonnie

12-29-2003, 11:06 PM

What channel? Can you double check the pigtails? We found out, the hard way, that you must push them straight down and the click as they go on. If you have the unit twisted and roll the connector you will not hear the click and when you examine the U.FL end you will only see one pin, since the other one has been rolled over and down under the plastic shell.

datawork

12-30-2003, 02:20 AM

Channel is 157 > 5875 have tried all in the 5.7 - 5.8 range with same result. Will check pigtails tomorrow as well as swapping out a Netgear PCI> Mini-PCI adapter that we put a Seano card in. I am somewhat suspicious of the Netgear adapters from past motherboard issues.

datawork

12-30-2003, 03:47 AM

Here's a wierd little bit of data.

Changed the ACK from 300 on the client end to 500 and viola all of sudden we are seeing

RX:2300KBps
TX:1300KBps

from the AP side (Dlink-ag520)

and 2000+ both ways from the client (Seano).

Going to swap out the D-link tomorrow witha Seneo card and see if we can get things up a little higher.

It would appear that the ACKs need to be synced on both ends for things to work corectly.

Will keep you posted, still waiting for the Equinox 32 dbi grids which we'll be testing on a 10 mile link.

Nice work Tony and Lonny

BTW the above was tested with build 13.3b10 on both ends with Intel p4 1.8 and one of your 733 VIA boards

tony

12-30-2003, 08:23 AM

I am glad to see your links is working. To gain extra performance, you may wish to continue increasing the ack value, maybe even to 1000 and see if your throughput increases.

datawork

12-31-2003, 01:46 AM

Bumped the ACK up to 750 on both ends any higher and throughput drops or the client stops associating.

Now seeing a steady 3000KBps TX RX and 6000 for both in Turbo Mode. :wink:

tony

12-31-2003, 09:04 AM

great news, thanks!

dkii

01-23-2004, 03:59 PM

I brought my first 5.8ghz link online today. One end is a soekris 4511(soon to be replaced by a celeron 2.4ghz) with an atheros a/b/g card, u.fl to n-female bulkhead pigtail, 3' lmr-195 NM to NM jumper, and a pac wireless 29db dish all purchased from wisp-router.com, mounted at about 75' AGL. Other end is a Routerboard 230, another atheros a/b/g card, I soldered an lmr100 n-male pigtail on this end and noticed about a 3db increase in signal over the u.fl pigtail. 21db superpass 5.1-5.8ghz panel antenna, polyphaser LSX arrestor, and a 10' NM to NM jumper. This end is 7.7 miles from the other end, and 485' AGL. Signals are around -74, when the speed is set to auto, it hovers between 18 and 24mb/s. I currently have it set to 18mb fixed. Speeds are about 1000KB/s using the internal speed test(Keep in mind one end is a 486 100mhz) I can't test the through traffic, because the soekris is connected to a 10base-t switch, so the through traffic won't be any better.

dkii

01-23-2004, 04:00 PM

btw ack is set to 9000. I tried backing it to 4500, 5500, and 6500, but could not get it to associate.

datawork

01-23-2004, 06:42 PM

Our Equinox 32 dbi grid finally arrived and put the first one up today on our 6.2 mile link. We where going to use these on our 10 mile shot but decided to make the 6.2 bullet proof.

Anyhow, with the Equinox now in place and a Pac Wireless on the other. The equinox end of the link has jumped to 25 on the quality and -70 for the recieve. Speed is jumping between 36 and 48.

Going to swap out the Pac Wireless tomorrow and see what we get.

BTW

Would not recommed the Equinox for windy areas without a side brace. The mount is similar to the Andrew 2.4 24 dbi Grid. The PacWireless 5.8 mounts are far superior and alot easier to tweek.

Mark

01-24-2004, 07:51 PM

it is a typographical error.

the grids are NOT that high of gain. Please call ecomm to verify which part numbers are what gain.

Thanks

datawork

01-25-2004, 12:42 AM

The 39" Equinox grids are 32 dBi, I have talked to Equinox directly about this.

Please refer to my post regarding the misprint in the Electro-comm catalogue

http://forums.star-os.com/viewtopic.php?p=9875#9875

datawork

01-26-2004, 03:04 AM

fired up second 32 dbi equinox grid today on our 6.2 mile link.

Still need to fine tune the dishes which is major pain in a** due to the mounts. One other thing we tryied today was comparing using D-link ag520's at each end and Seano a/b/g cards at each end. The results were somewhat surprising. Dispite the Seano cards being higher power (3dbm)m the throughput was lower than the d-link by almost 10 mbps.

Link info

AP 733 VIA C3

Equinox 39" 32 dbi grid
15' lmr 400
D-link ag520
Quality:25
Signal:-70 to -74

client 1.8 Intel P4

Equinox 39" 32 dbi grid
20' lmr 400
D-link ag520

Quality:27
Signal:-68 to -70

Ack 485 at both ends.

Star-OS throughput tests average 4500 KBps in both RX and TX from both client and AP with one exception.

The thoughput test done on the client in RX is consistantly pulling 5300 KBps with the peak being almost 5500 KBps (5487).

Mark

01-26-2004, 09:28 AM

That signal level is about 10 db lower than what you'd expect calculating path and device losses.

Any ideas why the discrepancy?

bobbyc

01-26-2004, 01:20 PM

The current hypothesis (found in some other thread) is that the atheros cards use OFDM when using .a and .g, and that the signal is probably there, but is not reported correctly.
On the other hand, people just don't realize how much signal is actually lost using lmr-400 and N connectors at 5.xghz... When I do my first distance shot, my radio pigtail is going directly into the antenna (radio mounted on back of antenna without even polyphasers)
Bob C

Mark

01-26-2004, 01:48 PM

well, I allowed 4db each for losses.

Do you think it should be higher? I don't plan on longer than 3 or 5 foot pigtails when I put stuff up at 5.8, so should it be planned on bigger losses? 5, 7 db?

Thanks

bobbyc

01-26-2004, 01:54 PM

I've never measured lmr-400 crimp on N-connector losses myself, but I've heard from a reliable source that 1.5dB is the max loss PER connector at 2.4ghz.. so if that is true then it would be even higher at 5.xghz. Standard 12" pigtail at 2.4ghz was like 2-3dB loss.
Times Microwave doesn't take into account the connnector(s) loss when giving you the attenuation of their cable per foot. Neither does the calculations... thus the 4dB that you put in for loss.
Of course my reliable source might have been telling me those loss figures just to make his calculations for his PTMP setup equal 36dB :oops:
Here's a post from a while ago about the connector loss at 2.4ghz.
http://forums.star-os.com/viewtopic.php?t=1239&highlight=cispa
I wish I still had the origional cispa email where I got the info... perhaps they have a archive.
Bob C

h3ll

01-26-2004, 02:46 PM

my observations on one of my links show that you should never use AUX port on atheros cards. always go to main. even you solder the pigtail to the card - go to MAIN. the difference is like ~5-10dB still dont know why.

go.fast

01-26-2004, 08:45 PM