Page 1 of 1

Short link Max throughput. Nstreme the answer?

Posted: Sat Oct 02, 2004 8:19 am
by dwright
Hello All,
Have a customer that needs to connect two building about 400' apart and needs a total throughput of about 50megs. Everyone posting seems to talk about these long links and nstreme doing these great bandwidth numbers, but what about short links. I wondering from those of you who have links pushing this much data, what equipment and drivers(nstreme, legacy), is needed to sustain this much bandwidth? Will a routerboard pass this much traffic, or is something more powerful needed? Any information would be a great help.

Thanks.

Dan

Re: Short link Max throughput. Nstreme the answer?

Posted: Sat Oct 02, 2004 10:38 am
by mag
if "50megs" means 50mbit/s half-duplex, this could be done with one nstreme link (802.11a-turbo, perhaps with 802.11g-turbo too), but more powerful hardware than routerboard is needed.

with dual nstreme 50mbit/s full-duplex should be possible.

regards.
matthias

Posted: Sat Oct 02, 2004 11:21 am
by stephenpatrick
Latest test results here at short range:
72Mbps sustained throughput half-duplex using MT bandwidth tester
- Single AR5212 or 5211 radio @5.8GHz
- 5GHz "turbo mode"
- Nstreme enabled
- 1GHz CPU utilisation ~60%
(figures strongly suggest a 266MHz board won't do that throughput)

Next tests will include dual-radio Nstreme-dual which should offer full duplex - note that needs 2 antennas though.

We're going to be packaging all this shortly into a "shippable product" with outdoor-grade housing for CPU, cooling, radios, N connectors to 1 or 2 antennas, outdoor-grade Cat5 cable feed to network, and pole/wall mounting brackets.
Using Desktop-PC chassis is messy and not what many customers expect for a wireless installation. I Intend to post some pictures when it's all together.

(For such a short distance, have you considered a Laser link? at such short distances, you can get 100Mbps or Gigabit Ethernet full duplex. To avoid any inappropriate product pitch on the MT forum, do contact me off-line - stephen.patrick@cablefree.co.uk)

Regards

Stephen

Posted: Sat Oct 02, 2004 10:09 pm
by tully
We have tested more than 84Mb/s throughput with this config -- I think it gets close to 90Mb/s, it depends on antennas (this is on the lab table).

The CPU usage is from the bandwidth test running on the same box -- most likely, you didn't say where it was run from.

John

Posted: Sun Oct 03, 2004 8:18 am
by dwright
John,
What kind of hardware have you tested with these numbers? I understand that running the bandwidth test from the board will add cpu utilization. Just wondering how much cpu would be needed to run those numbers? We are looking to get 50Mb/s half duplex, but would like room to grow, possibly adding a second radio to have full duplex in the future if the need arises. Although having 84Mb/s or even 90 would be better.

Dan

Posted: Sun Oct 03, 2004 8:45 am
by tully
Celeron minitowers.

John

Posted: Sun Oct 03, 2004 11:50 pm
by goldclick
Latest test results here at short range:
72Mbps sustained throughput half-duplex using MT bandwidth tester
- Single AR5212 or 5211 radio @5.8GHz
- 5GHz "turbo mode"
- Nstreme enabled
- 1GHz CPU utilisation ~60%
(figures strongly suggest a 266MHz board won't do that throughput)

Next tests will include dual-radio Nstreme-dual which should offer full duplex - note that needs 2 antennas though.

We're going to be packaging all this shortly into a "shippable product" with outdoor-grade housing for CPU, cooling, radios, N connectors to 1 or 2 antennas, outdoor-grade Cat5 cable feed to network, and pole/wall mounting brackets.
Using Desktop-PC chassis is messy and not what many customers expect for a wireless installation. I Intend to post some pictures when it's all together.

(For such a short distance, have you considered a Laser link? at such short distances, you can get 100Mbps or Gigabit Ethernet full duplex. To avoid any inappropriate product pitch on the MT forum, do contact me off-line - stephen.patrick@cablefree.co.uk)

Regards

Stephen
Stephen,

Am really interested in knowing what distance you achieved the 72mbs half duplex. Am currently setting up a 15km link using nstream dual with two Ath5212 cards in 2.4GHz minitower PC to carry 50mbps duplex. Am willing to setup another nstreme dual repeater should this distance be too large to achieve this speed. Any ideas and experience will be appreciated. Thanks.

Sonny

Posted: Mon Oct 04, 2004 12:26 am
by stephenpatrick
Hi Sonny,
Our tests were "lab tests", antennas only a few m apart.
We know there will be some degradation with distance, but we'll have to do field tests to check that.
Some of the other users on the forum have quoted 50Mbps half duplex at 35km which is quite phenomenal.

John:
Great to hear you can get even more than 72Mbps- impressed!
We were running bandwidth tests "through" the box using external PCs to generate the traffic.

Do you have any sort of "chart" predicting throughput drop-off vs. distance using Nstreme?
Even a few "spot figures" would be really helpful.
(and for marketing, a comparison without Nstreme or "vanilla" 802.11x)

Our focus is getting the solution running well with small, cool-running motherboards in external-grade packages that will tower-mount and survive all weather conditions. That way, the RF cables (and losses) will be short. CPU heat dissipation at these rates seems to be a key issue.

Stephen

Posted: Mon Oct 04, 2004 11:12 pm
by tully
The only drop off in speed is when frames are lost and have to be resent.

This would be a factor of:
- signal strength
- multipath problems (signal bouncing and canceling another signal)
- interference

John

Posted: Mon Oct 04, 2004 11:34 pm
by stephenpatrick
Thanks John!

I think a gauntlet has been thrown down -

Who can post some results on a "real" outdoor link, with more than 70Mbps throughput - longest range the better ?

That would give some challenge to the "marketing supporters" of the WIMAX camp also ...

Stephen

Posted: Tue Oct 05, 2004 4:17 pm
by daniel
Hi Sonny,
Some of the other users on the forum have quoted 50Mbps half duplex at 35km which is quite phenomenal.
Here are more 'precise' numbers. Please note that this test is done on a live link, that already carried over 10 Mbps Internet traffic (that is, not large packet traffic). The setup is pretty much generic, wihtout


status: running
duration: 2m18s
rx-current: 47.5Mbps
rx-10-second-average: 45.8Mbps
rx-total-average: 44.5Mbps
lost-packets: 515

ping at the same time shows

126 packets transmitted, 124 packets received, 1% packet loss
round-trip min/avg/max = 2/15.2/35 ms

CPU utilization is ~95% on the sending (remote) router and 70% on the receiving (local) router. tests are run between both routers only.

Without the test, CPU load is 40-50%

Distance is 35km, antennas 1.4 solid parabolic, 5G/ABM modules at both ends, dual-nstreme, 5GHz-turbo

Bidirectional is more impresive:

. status: running
duration: 54s
tx-current: 22.2Mbps
tx-10-second-average: 21.6Mbps
tx-total-average: 19.2Mbps
rx-current: 45.9Mbps
rx-10-second-average: 45.3Mbps
rx-total-average: 37.6Mbps

CPU utilization is 100% on the local router and 95% on the remote router.

Daniel

Posted: Tue Oct 05, 2004 4:20 pm
by daniel
Our focus is getting the solution running well with small, cool-running motherboards in external-grade packages that will tower-mount and survive all weather conditions. That way, the RF cables (and losses) will be short. CPU heat dissipation at these rates seems to be a key issue.
Stephen, were you able to find anything over 1 GHz to run at low power (temperature :). It seems for singe/dual links an 1GHz C3 is sufficient, but that makes the number of boxes on a repeater site large.

Daniel

Posted: Tue Oct 05, 2004 7:49 pm
by stephenpatrick
Hi Daniel,

Yes we have made a huge strides in the last few days.
The EPIA V10000's never worked reliably, even with a "patched" beta bios - as John Tully pointed out, the Atheros cards fight with the ethernet on the PCI bus, we were getting bursts up to 10Mbps but nothing sustained.
We have some EPIA M10000's which still need more testing, but certainly have enough CPU power and run cool.
We also have an Intel Pentium 'Mobile' solution on test which is excellent and cool-running too, though more expensive.
In this case we can get 2 radios, which would be excellent for a repeater site.
So our focus is to get these tested into "known good" condition and integrated into high-quality external-grade boxes. That's going to be very soon (days).

BTW, who do you work for in Bulgaria? e-mail me off-list if you want:
stephen.patrick@cablefree.co.uk

Regards

Stephen

Posted: Wed Oct 06, 2004 1:19 am
by goldclick
Hello Daniel,

Thanks for posting the test results. It's impressive! What grade and length of cable do you have between the 5G/ABM and the antennas? Any amplifier?

Sonny