Community discussions

MUM Europe 2020
 
hochtr
just joined
Topic Author
Posts: 14
Joined: Fri May 28, 2004 10:20 pm
Location: Austria - Vienna
Contact:

Nstream testing

Tue Sep 14, 2004 10:29 am

PtP-Link: Distance is 8 km both 2.8.16 2,4 Ghz Celeron and 1 Ghz Celeron

Signal TX and RX about 68-70

10 Mbit in one direction without nstream isn't problem ( scp from linux to linux)
with nstream the performance is about 7 mbit

The Problem is that on "Station-Box" the TX-Level change always between 6 and 36 Mbit - without nstream the Level is constant 36 Mbit.

There is no difference with changing some settings (polling, fit size, paket size, ..)
Pine Networks
Cable and Wireless
 
tully
MikroTik Support
MikroTik Support
Posts: 505
Joined: Fri May 28, 2004 11:07 am

Tue Sep 14, 2004 11:52 am

If you can send us the supout from both routers when Nstreme is enabled, we can check them.

John
 
mp3turbo2
Member Candidate
Member Candidate
Posts: 196
Joined: Wed Jun 02, 2004 9:15 am

Tue Sep 14, 2004 12:41 pm

I would guess that this fluctuation is more card-dependent than software dependent. We have replaced some noname taiwan cards with brand-name ones and the link has trippled in throughput and totally stabilized pings (1.5ms to 2.5ms) as the radios do not switch speeds every second.

Of course, there could be some nstreme issue... But I have report of _excellent_ nstreme performance and stability with 2.8.13, 2.8.15, right now those people are testing 2.8.16. When I say excellent, I mean excellent. I have never seen such numbers here in forum nor everywhere on internet.
 
tully
MikroTik Support
MikroTik Support
Posts: 505
Joined: Fri May 28, 2004 11:07 am

Tue Sep 14, 2004 2:37 pm

Let us know the numbers when you have them.

John
 
User avatar
[ASM]
Member Candidate
Member Candidate
Posts: 285
Joined: Sun Jun 06, 2004 12:59 am
Location: Sofia, Bulgaria
Contact:

Tue Sep 14, 2004 8:26 pm

btw... when updated to 2.8.16 the max speed droped from 44 MB/s to about 32 MB/s... after downgrade to 2.8.15 speed was 44 MB/s again
 
hochtr
just joined
Topic Author
Posts: 14
Joined: Fri May 28, 2004 10:20 pm
Location: Austria - Vienna
Contact:

Wed Sep 15, 2004 10:40 am

[quote="mp3turbo2"]I would guess that this fluctuation is more card-dependent than software dependent. We have replaced some noname taiwan cards with brand-name ones and the link has trippled in throughput and totally stabilized pings (1.5ms to 2.5ms) as the radios do not switch speeds every second.

That is not a "labor test", it's a test under bad condition - I know.
There could be many problems (mtu size, two cards in one box, bridge, eoip tunnel, scp with ssh..) but for us is this a test to simulate the customer-speed.

The speed is not the problem but the differenz between with nstream and without.
Without nstream the link is very stable (36 mbit modus constant, ping 1 - 2 ms ).
Maybe it is depend from the hardware.
Pine Networks
Cable and Wireless
 
tully
MikroTik Support
MikroTik Support
Posts: 505
Joined: Fri May 28, 2004 11:07 am

Wed Sep 15, 2004 1:20 pm

To get true throughput, you have to use UDP, maybe your techs tested in different ways. Also, make sure the random-data is set to 'no' -- I think this is the default in all newer versions.

John
 
hochtr
just joined
Topic Author
Posts: 14
Joined: Fri May 28, 2004 10:20 pm
Location: Austria - Vienna
Contact:

Wed Sep 15, 2004 2:57 pm

further tests with the bandwidth-tools:

udp in one direction about 23 Mbit
tcp in one direction about 20 Mbit

udp in both direction about 12 -13 Mbit

If there is much traffic over the link then the radios dont switch down to 6 Mbit. If the load on the link is low then the radios switch always between 6 and 36 Mbit and the pings are not so good (4 - 12 ms ) as without nstream (1-2 ms).
Pine Networks
Cable and Wireless
 
tully
MikroTik Support
MikroTik Support
Posts: 505
Joined: Fri May 28, 2004 11:07 am

Wed Sep 15, 2004 3:53 pm

I think there are hardly any changes between .15 and .16 (or none) for nstreme -- so we will have to think a little on this.

John
 
daniel
newbie
Posts: 47
Joined: Sat May 29, 2004 11:46 am
Location: Bulgaria

Fri Sep 17, 2004 8:17 am

I think there are hardly any changes between .15 and .16 (or none) for nstreme -- so we will have to think a little on this.

John
John,

My view on this is, that the nstreme protocol tries to switch speeds between send/receive in order to accomodate the different traffic allocation in each direction. (I may be wrong :)

Another possible variant is that the 'traditional' (that is the non-nstreme protocol's) algorithm for speed shifting is used, which may not be appropriate for nstreme.

What happens in real world is, that the speed in the direction at which most traffic flows is always higher, and the 'return' is shifting speeds - when there is no traffic, the speed is 6 Mbps, when there is traffic the speed rises. These shifts are what is causing the extreme RTT variations with nstreme and thus the poor performance.

Earlier versions of nstreme did not exibit this behavior. I can never reproduce the 35 Mbps on 57km, with recent nstreme revisions.

Daniel
 
mp3turbo2
Member Candidate
Member Candidate
Posts: 196
Joined: Wed Jun 02, 2004 9:15 am

Fri Sep 17, 2004 9:40 am

hello folks,

> PtP-Link: Distance is 8 km both 2.8.16 2,4 Ghz Celeron and 1 Ghz Celeron
> Signal TX and RX about 68-70

ok, let's go : I consider -70dB signal on 8km link to be kinda weak, it could be stronger. As you probably know, there are three things that count with OFDM modulation:

a) signal strength, every decibel counts (!!)
b) interference on given frequencies - definitely this is MAJOR factor of performance
c) directionality of signal - OFDM does not like sector / panel antennas, PtP links are ideal


Next thing: try locking your nstreme settings for 36Mbit speed:

/interface wireless
print
set name-of-interface supported-rates-a/g=6,9,12,18,24,36Mbps
set rate-set=configured (!!!!!!) - this will LOCK your setting to 36Mbit max.

If you want to lock your card to 24Mbit and 36Mbit datarate only, delete 6,9,12,18Mbps figures from supported-rates-a/g setting - but beware that if you don't have enough signal to hold 24Mbit connection, the link will drop and won't transmit traffic AT ALL. Don't forget to change your basic-rates-a/g to 24Mbit/s also.

Yes, could be there is different behaviour between legacy and nstreme drivers - legacy holds your 36Mbit settings whereas nstreme could be 'speculating' or 'behaving' little bit differently. Lock it with above settings and there you go.

Also, I feel that your performance could be little bit higher for the signal strength you have. Are you sure there are no 'cleaner' frequencies in your band ? Did you try another freq ?

Let me know if my setting did help you to lock cards to 36Mbit.

bye, mp3turbo.
 
tully
MikroTik Support
MikroTik Support
Posts: 505
Joined: Fri May 28, 2004 11:07 am

Fri Sep 17, 2004 11:55 am

Daniel,

If you can give us access to both of the 57KM routers, we can see what the problem is. The Nstreme protocol is quite complicated, so I it would be difficult to explain it all and also I don't want give help to competition. Of course from your test results, it seems we need to do some more test.

John
 
hochtr
just joined
Topic Author
Posts: 14
Joined: Fri May 28, 2004 10:20 pm
Location: Austria - Vienna
Contact:

Fri Sep 17, 2004 4:24 pm

>ok, let's go : I consider -70dB signal on 8km link to be kinda weak, it >could be stronger. As you probably know, there are three things that >count with OFDM modulation:

tx-power is set to 14 on box A and 16 on the second box


>b) interference on given frequencies - definitely this is MAJOR factor of >performance
I am sure that is not

>c) directionality of signal - OFDM does not like sector / panel antennas, >PtP links are ideal
there are two parabol antennas (29 db + 10 meter cable and 24 db + 2 meter).
I locked the radios to 36 Mbit but without success.
The only problem is the ping-time ( 4 - 16 ms) - the performance (mbit) is not the problem.

With the legacy package the link is very stable - 3000 pings without package loss - what is the difference ?
Pine Networks
Cable and Wireless
 
mp3turbo2
Member Candidate
Member Candidate
Posts: 196
Joined: Wed Jun 02, 2004 9:15 am

Sun Sep 19, 2004 1:22 am

> tx-power is set to 14 on box A and 16 on the second box

do NOT use tx-power setting, we always got lower throughput when we set anything other than DEFAULT, so card decided value. Set tx-power=default !

> I locked the radios to 36 Mbit but without success.

ouch.

> The only problem is the ping-time ( 4 - 16 ms) - the performance (mbit) is not the problem.

what pings do you have with legacy?

> With the legacy package the link is very stable - 3000 pings without package loss - what is the difference ?

more stable wireless driver without tries to implement own wireless protocol - there is some tuning necessary, although we have to admin that nstreme works and works beautifully. Just not always :)

bye, mp3turbo.
 
hochtr
just joined
Topic Author
Posts: 14
Joined: Fri May 28, 2004 10:20 pm
Location: Austria - Vienna
Contact:

Mon Sep 20, 2004 11:46 am

>do NOT use tx-power setting, we always got lower throughput when we >set anything other than DEFAULT, so card decided value. Set >tx-power=default !
I am not agree - tx-power is a standard feature and in some situation is it better to decrease the power (mostly better SNR - not overpowered the second amplifier ) and a signal strength between 68 - 70 is good enough for 36 Mbit

>what pings do you have with legacy?
with legacy: 1- 2 ms very stable

my conclusion:
under heavy load ( over 10 Mbit ) nstream works better
under low load lagacy works better for us
so why not mix and change the protocol automatic...

Kind regards,
Horst
Pine Networks
Cable and Wireless

Who is online

Users browsing this forum: amt, blackmetal, ipaulponce, jonhyg, mrmut, OlofL and 179 guests