2 Port WAR Board

[lonnie]

We have arrived at the board we will use for the 2 port system. The unit has a 266 MHz IXP-420 processor, 2 mini PCI slots, 2 Ethernet, 32 MB ram, 4 MB flash, some nice LED lights and it takes an input voltage of from 12V to 48V. It draws a nice 6W with two CM9 radios.

It will be available for evaluation in about 4 weeks and in volume in 6 weeks. Pricing is pretty decent for the first order and will drop as our volume increases. We will be very aggressive on larger orders. This is the next generation and we are so pleased to be offering it.

[cyprus]

The prices for these lower models are a lot more affordable, however, what is the achievable throughput with these? Is there noticable difference with WRAP board's 20mbps?

[mp3turbo]

it was mentioned many times here in forum - yes, they are capable of full atheros throughput, thus ethernet could be also fully utilized. You can't compare this by any means with old wraps, they'd look like crap :)

Lonnie, congrats! I have the feeling that new era began. Only if I had x86-v3 in my hands... :)

bye, mp3turbo.

[lonnie]

Since the v3 will share so much of the code base from the Vx we really are developing it as we go. It will not have the performance but will have the features PLUS since it is being developed for much larger machines. The Vx is for small non x86 machines with very distinct hardware. And don't worry, there will be a v3 release for the WRAP board.

We can drive the Atheros to full speed in Turbo mode with the 2 port, and that is over 6 MBytes/sec non compressed transferring a zip file. With compression this goes up, but of course depends on the material being transferred.

Yes, I think we have kicked off a new era. This product is not just another repackaging of freely available code for off the shelf systems. Most of it is our own code. It will evolve as we add features. It has taken a long time to get to this point, but that is always required with a new foundation. Once that first above ground level goes up the building seems to take shape rather quickly.

This next year is going to see a steady roll out of products as we continue to push forward.

[mp3turbo]

thank you for comment!


I wouldn't like to turn this thread into v3 Q&A, so let's stick to 2port WAR topic right after my question :) do you have any estimations on speed degradation when using standard x86 platform or any preliminary speed tests? Ie:

WAR 266MHz : atheros can do 3.5MB/s with zip files, no more, cpu is limit
WAR 400MHz : atheros can do 6MB/s, full throttle, cpu is no longer limitation
x86 400MHz : atheros can do 5MB/s approx (maybe 4.7, maybe 5.2), cpu is limit
x86 600MHz : atheros can do 6MB/s, full throttle, cpu is no longer limitation
x86 1000MHz: atheros can do 11MB/s :)

thnx, mp3turbo.

[lonnie]

WAR 266MHz : atheros can do 6.20MB/s bridged with zip files, cpu is limit
WAR 533MHz : atheros can do 10.7MB/s bridged, full throttle, cpu is no longer limitation
x86 400MHz : atheros can do 7MB/s approx (maybe 4.7, maybe 5.2), cpu is limit
x86 600MHz : atheros can do 11MB/s, full throttle, cpu is no longer limitation
x86 1000MHz: atheros can do 11MB/s, lots of time left for routing

[tony]

I have done some recent testing, with roughly these results. (testing was done on the bench, with -54dbm signal, and little to no interference)

WAR testing results in 802.11a TURBO, QoS disabled, all super features enabled:
* WAR 266Mhz: (processor is at capacity)
BRIDGED: 6.70MB/sec download of compressed .zip file (atheros->ethernet)
BRIDGED: 10.7MB/sec download of zero-filled text file (atheros->ethernet)
ROUTED: untested at this time

* WAR 533Mhz:
BRIDGED: 6.70 - 7.00MB/sec download of compressed .zip file (atheros->ethernet)
BRIDGED: 11.50MB/sec download of zero-filled text file (atheros->ethernet)
ROUTED: 6.70MB/sec of compressed .zip file (atheros->ethernet)
ROUTED: 10.7MB/sec download of zero-filled text file (atheros->ethernet)

Both systems get 11.50MB/sec ethernet->ethernet, routed or bridged.

[isabido]

Is the bridge configuration test?


LAN---(ethernet-WAR-atheros)--->>> wireless link <<<----(etheros-WAR-athernet)---LAN

[mp3turbo]

I'll write something (rather clever) when I'm back from hospital.


I got a heart attack.

[butchkemper]

Tony,

Would it be possible for you to repeat the tests using just the 802.11a mode without the Turbo option?

Thanks.

Butch

[tony]

I will get the 802.11a non-turbo results posted a little later this morning.

The test configuration is like so:

Windows XP w/ Atheros linked to WAR AP, and off the WAR AP's ethenret is a 1Ghz StarOS box, and a webserver nearby (local LAN) with http test data.

To test the zero-filled data, we use starutil, and the .zip transfer is from the webserver using various different download utilities.

It is not scientific, but suites our purpose during testing.

[tony]

Updated results for non-turbo mode on a 533 (should be similar on the 266Mhz platform)

WAR testing results in 802.11a, QoS disabled, XR disabled, all super features enabled:
* WAR 533Mhz:
BRIDGED: 3.50MB/sec download of compressed .zip file (atheros->ethernet)
BRIDGED: 8.90MB/sec download of zero-filled text file (atheros->ethernet)

[isabido]

Tony,

This test is short distance, no?

You have some test long link, over 20km?

The vxWorks, have a propietary protocol (example,nstreme mikrotik) for long distance link? only correct the ack timeout?

Thanks in advanced!

PD: Sorry, for my poor english.

[lonnie]

This test is short distance but longer distance will be fine. The key is signal strength and ack timing. We saw this same correlation with the WRAP and Linux, so we are quite confident about our numbers at farther distances.

[lonnie]

I'll write something (rather clever) when I'm back from hospital.


I got a heart attack.

How are you doing? I was hoping you would report how things are and not just dive in as if nothing happened.

[mp3turbo]

> How are you doing? I was hoping you would report how things are and not just dive in as if nothing happened.


thanks for asking, I'm very fine - that heart attack was just fake because of speed you mentioned. I apologize if I did any harm. Yes, I have yet to see 7.5MB/s real transfer of zipped file thru wireless.

All in all, the best sign that I'm doing pretty well was my bonding speculation in another thread :)


PS : a friend of mine was doing his pilot training for ultra-lights when his old, ca. 65yr instructor pretended a heart attack. Senior let the young rabbit to cope with that situation for ca. 20 seconds until he suddenly "came back". I can't reproduce what senior said that my friend said :)

bye, mp3turbo

[rbolduc]

How are you doing? I was hoping you would report how things are and not just dive in as if nothing happened.


Yea I thought the same but after reading the thread again it sounded more like you gave him a heart attack with the speeds you posted..


Reed

[rbolduc]

> I can't reproduce what senior said that my friend said :)

bye, mp3turbo

Umm something along ^%** *&^% ^&&* *&^%$#@ &^%$*@# mabye a little longer and in a different order..

LOL
Reed

[phendry]

This test is short distance but longer distance will be fine. The key is signal strength and ack timing. We saw this same correlation with the WRAP and Linux, so we are quite confident about our numbers at farther distances.

For the WRAP's we needed a signal quality of around 20 to give full and stable (CPU restricted) throughput. What would you say is the minimum signal quality to get full throughput on a none "turbo" WAR system?

[lonnie]

It will be the same. The cards require a -70 dB or better signal to hit 54 mbps air rate, and there is no driver magic that will change those rules of radio communication.

[Mark]

It will be the same. The cards require a -70 dB or better signal to hit 54 mbps air rate, and there is no driver magic that will change those rules of radio communication.

Just a question about this...

I've got some links that run real close to this... -69 or so...reported... and they'll only work at 36 - 48 will connect, but won't communicate effectively.

How much better than -70 do you ahve to be?

[lonnie]

Sorry, I don't know. Up until now it was not a factor since we did not have the processor to take advantage of it. We set speed to auto and it just moves up and down as required. Mostly we see about 36 to 48 with -70 dB signals.

[stephenpatrick]

Has anyone taken this solution "outdoors" and tried at range?

Also the question before, "how does it compare with Nstreme"- that is important because AFAIK Nstreme encapsulates smaller packets to make "jumbo packets" which makes the link faster when passing "real world" traffic.

[lonnie]

We have a link at 81.5 km that gets -80 dB signals with 3 foot dishes. We are only seeing about 6 mbps on this link.

We also have a link at 23.5 km using the little SuperPass 21 dB patches, also getting about -80 dB signals. We are getting about 20 mbps on this link.

Our testing is leaning towards distance and minimum antenna sizes to maintain a link. In general, for more than 90% of ISP customers, the ability to deliver more than 5 mbps is adequate, so if you can do a 23 km shot with $60 antennas it is good enough.

I have no idea about Nstreme. I do know that our driver was one of the first to take advantage of the Atheros advanced features, including compression, packet aggregation (jumbo packets) and bursting (sending more than 1 packet at a time).

In order to get high data throughput you require LOTS of signal, and in that respect Nstreme is not better or no worse than our methods.

You will always be better of in taking several smaller hops with high signals. They are easier to do and have the ability to pick up susbscribers in the vicinity of the repeater. It is also one step to a mesh.

[stephenpatrick]

Hi Lonnie,

Thanks for the reply. I understand your points, guess most of your users are WISPs. We also have campus and corporate users who want max speeds out of the equipment and will pay for performance and reliability. Often they can't put up repeater sites, hop lengths are determined by the building locations.
Without stirring up some "vendor competition" I would love to see some "real world" comparison with MT/Nstreme. It sounds like your driver support is very advanced. I like the idea of the IXP425 (WAR) platform as the cost, performance and heat generation are definitely on it's side compared to x86 platforms.

Yesterday we tested MT on 7.2km over water with 23dBi flat panels, 5.8GHz. We got 40Mbps through sustained UDP using "bandwidth test", limited by the signal strength (-71dB) which meant 24Mbps modulation, in turbo mode. (With decent signals >-60dBm we get 60-80Mbps through). It would have been fascinating to know how your solution would have compared.
Regards

[lonnie]

I'm curious to know the hardware that you are using for the 40 mbps and the 60-80 mbps speed tests.

Things get murky when comparing apples and oranges. The only way to compare the two would be in actual use, ie a real live install and test using a large file ftp transfer.

We both use different speed tests and the numbers are not comparable. Our test is comparable within "our" units since they are all relative to the same method, but you cannot take a 40 mbps udp transfer on MT and compare to a 40 mbps tcp transfer on StarVx.

Currently we have a shot that is about 1 km but we have it as a repeater, meaning that the traffic goes through 4 radios on its way between the systems. In this test we can pull a fairly consistent 3.5 MBytes/sec of TCP traffic. This is routed and we have not yet tested with bridging since we were concentrating on getting an update version with RIPv2, and it was just released today. The next few week will see some real big changes and additions.

[stephenpatrick]

Hi all,

We achieve 60Mbps UDP using a 600MHz Pentium M platform,and 82Mbps using a 1GHz platorm. That's using Mikrotik bandwith test, through the router, in the lab.

The 40Mbps result was a "live" test over 7.2km, over water with partial fresnel zone obscuration, using 600Mhz platform and 23dBi antennas. The link budget was limiting, i.e. -72dBm received level, modulation set to 24Mbps turbo mode, so we know that with better antennas at full signal level we could have achieved the full throughput, i.e. up to 82Mbps.

I totally agree about apples and oranges, there needs to be some standardisation of tests - UDP, TCP, FTP, setup, etc.

Look forward to hearing more -

Regards

[tony]

In the lab testing one of our IXP420 / 533Mhz systems, we can obtain ~10,500 KBytes/sec (82Mbps) raw throughput using TCP.

Setup is something like this:
2Ghz Windows XP (Atheros) --( ~~ )-- 533Mhz WAR4 AP --x-- 1Ghz StarOS box

The Windows XP system is doing a starutil throughput test (TCP) to the StarOS system.

The WAR AP is set to 11a TURBO, QoS disabled, Super A/G enabled.

The Windows XP system is associated to one of the WAR4's four cards, the StarOS box is on the same LAN segment as the WAR4's ethernet, and is acting as the throughput test server.

The same setup, but using standard 802.11a mode (non-turbo), no QoS and SuperA/G enabled, can get roughly 9,000KBytes/sec (70.3Mbps).

We have done some preliminary testing with UDP, and found the throughput to be generally higher, though we do not have any numbers at this time.

Additional note: The WAR board's association display shows the throughput of each associated station, which is very accurate, and is what we use as a guide.

[tony]

The 2-port board will be running at 266Mhz, however testing shows it is able to easilly obtain 70Mbps.

[lonnie]

I might add this is pretty decent for a 4"x6" circuit board that holds 4 radios and draws less than 8W (for the 4 port board). This makes it dirt simple to tower mount and it is perfect for Solar Power repeater sites.

The 2 port board will have less ram, flash, smaller power supply and a 266 MHz processor. It will draw less than 5W and is the same size 4"x6" board.

All the numbers that Tony is quoting are TCP traffic throughput. We will work out a UDP test, but I fear the numbers will be quite a bit higher. Do you know if the MT tests are taking numbers from the sender or the receiver?

[phendry]

Stephen, how much cable loss did you have and what power output did you use on the cards?

Tony, I thought TURBO and SuperA/G where the same thing. Has StarVX got the option to enable either/or? If so, can you clarify the difference?

[tony]

Turbo and SuperA/G are not the same thing. Turbo is 40Mhz wide band, while non-turbo is 20Mhz.

SuperA/G option enables Atheros-specific features such as packet aggregation, compression and bursting which gives the throughput a good jump. To use SuperA/G, both the AP and client must be Atheros, with the features supported.

Update:

Turbo is classified as channel bonding, which is a seperate component from SuperA/G. In regards to 802.11g, it's turbo mode depends on SuperA/G features, and must be enabled for 11g turbo option to be available.

[stephenpatrick]

Hi all,

Tests were with CM9s in default power output - that's less than 65mW on the datasheet, and I suspect quite a bit lower than that in practice.
We were using 4-5m long RF cables (only ones we had lying about) but it's very thick coax so low loss (sorry, am not in office so can't check what spec it is) with N connectors both ends, and pigtails to the Atheros cards.

Tests were **through** the routers, measured at the testing terminal (here they were win2k PCs running MT windows bandwidth test.
The MT GUI showed about 40Mbps going through the interface, so the readout was showing true results of throughput.

As I pointed out, this was antenna/link budget limited, with better signal levels in the lab we get 82Mbps UDP or 72Mbps TCP using the test utility.
Very interested by the Atheros driver support you have - seems a few more features are available to switch on/off - and if you are right about 83Mbps TCP, you have better throughput, at least in lab test conditions.

Assuming I want to do some comparitive lab tests - do you have any 533MHz platforms (a pair of boards) in stock anywhere?

Regards

[lonnie]

There will be stock after the 15th of August for the 4 port boards.

[datawork]

Is there going to be a single WAR radio system available and if so when?

We're running out of wrap boards and based on the two radio WAR system. It will most likely be more cost effective for us to order the WAR boards for you rather than buy the WRAP's from one of the US based WRAP resellers.

I'd prefer to give you guys the business anyway.

Though I'm not thrilled about your decision to not stock the WRAP platform anymore. I do understand the reasons behind the decision and will continue to support your business if at all posible.

Keep up the good work.

[lonnie]

There will be a single port WAR, with a different design that is at the prototype stage. We cannot afford another line so it must wait for production. Maybe in the New Year.

As our volume builds we will be able to get the WAR to the WRAP pricing levels.

[datawork]

maybe I should rephrase the question

Will the 2 port WAR board be available with a single CM9 for the time being?

[lonnie]

Initially it will not be offered with a single radio. You might not like this, but we have to test with all ports by installing radios in each slot to make sure they work. It is a lot more work to the remove the radios. I am not sure how many times you can remove a radio before it can become a source of trouble and itself needs to be replaced.

The client system is coming. Please be patient.

[luka]

Are you planning to put in StarVX something like Nstreme option from Mikrotik? Two radios from war-2 may be used for rx/tx each separately. Suppose this may gain even better performance overall.

[lonnie]

We have plans for using two radios for a single link.

[David L. Vrablic]

And the client is comming too?
Wonderfull.

:grin: HORRAY!
Go Lonnie, Go Lonnie, Go Lonnie GO!

(And Tony)
Great News!

[lonnie]

We have about 100 of the 2 port WAR in stock. We will need 2 weeks of testing before they ship since we have not seen these before. They are different and will require a bit of modification to the image. They are getting real close. Let the fun begin.

The client prototype was received yesterday. It should be running in about a month, so that means it will be ready by mid to end of November for shipping. We thought we'd put them in Rootennas and sell as a package.

[phendry]

I take it there will be the option to buy the single port client without the Rootenna? We have developed our own CPE with built-in 19dB antenna (and they look a bit more pleasing to the eye ;))

[lonnie]

For a huge order we can do almost anything you want. ;)

[stephenpatrick]

Well I'd second phendry's post, the market seems to demand flexible solutions.
If you're going to sell a cut-down single-radio board, you need volume to make it viable, and you will find OEM customers wanting "bare boards" the ones who will give you the largest potential orders. No sense in restricting them -

Regards

[go.fast]

I take it there will be the option to buy the single port client without the Rootenna? We have developed our own CPE with built-in 19dB antenna (and they look a bit more pleasing to the eye ;))

I'd like to know what your using for an antenna enclosure.
If you had a url, I'd be very grateful.

I hope Lonnie offers the single port client boards seperate like he does now with the other boards. I need a choice for antennas.

George

[Tom Sharples]

Lonnie, your new WAR code sounds very exciting and I can't wait to test it. Is it available as a CF or downloadable binary so that we can use this board equipped with a pair of CM9's:

http://hollowst8.hypermart.net/images/ixp420.jpg

Thanks!

[lonnie]

All systems are different enough that we must tune to each platform. VxWorks is not a general purpose, works on anything system.

The Vx software will only be available on equipment that we provide. We have enough to do right now without supporting a whole bunch of other boards.

[tog]

I do hope the "client" WAR boards will be available without the rootenna as well.

I already find it restrictive that I cannot buy a WAR board without also buying a CM9 card in every slot with it. I have not yet had reason to buy more than a single WAR board for testing, so I have not yet asked you in private if you will make exceptions for people who do not need so many CM9s.

I'd rather not have any buying restrictions, such as only being able to buy the single-slot WAR client board with one CM9 and a rootenna. I might want to put it inside a higher-gain rootenna (or similar enclosure) or put it inside a weatherproof enclosure and connect an external antenna or even use a 5GHz rootenna instead of a 2.4GHz rootenna.

Only being able to buy bundles is especially bothersome if you are the sole source for the WAR boards.

[lonnie]

We will use the highest gain rootenna available.

We are considering a special license that will indicate how many radios to be activated. When we have that worked out we will be OK to sell with as few as one radio.

[go.fast]

Are you saying that you may not be selling boards only in the future?
I don't have too much of a problem pulling cards, but I can't imagine not being able to buy a 4 port board complete without enclosure, or antenna and firing up the amount of cards I want. I don't want to have to do a relicense to add a card, haven't got time for that.

Rootennas might be nice, I use quite a few of them now, but their usefulness is very limited, regardless of gain. I just tried one of the 5.8 pac wireless ones in a PtP war board link and was much happier when I used a superpass panel.

Don't know if you ever seen a 2.4 19db pacwireless rootenna.
It ain't small.

George

[lonnie]

We are NOT only going to be selling cards in rootennas.

You WILL be able to buy them as bare cards or in our choice of enclosure.

The plan is to have singles, doubles and quads, so you choose your product based on the number of cards you need.

[tog]

I don't think you generally want to use the highest-gain rootenna available.

Pac Wireless makes a 19dbi 2.4GHz rootenna and the thing is a huge ugly heavy BOAT. I could sail to Cuba in it.

[povalnik]

I take it there will be the option to buy the single port client without the Rootenna? We have developed our own CPE with built-in 19dB antenna (and they look a bit more pleasing to the eye ;))

Prompt as you it have made (if it not the classified information) or send me via e-mail cost your CPE.

[budi]

We're too have CPE with :
- Support 802.3af
- Builtin 20 dBi 5 Ghz antenna (5150 s/d 5850) Ghz
- Support ACK timeout adjustion/

Thank's
//Budi

[szern]

Any information you can post about your CPE Budi?