poor performance whit small packets !

cuccio · Wed Oct 11, 2006 8:41 pm

Wi find a big problem with small UDP packets.
wi try to send a 64byte (-l 64b) packet stream with Iperf and can't obtain an accetable througput. We have measured max 8.000 Packet/sec and maximum transfer at 6Mbps

.
Alvarion BrezzeAccess VL v.4 have 40.000 packets/sec and 20Mbps in the same condition

.
I used Iperf with following strings:
server: Iperf -s -u -i 1 -w 128k -l 64b
client: Iperf -c 192.168.x.x -i 1 -w 128k - l 64b -b 10m

where 10m is the stream, and can change to obtain minimum packet loss.
Alvarion support -b 20m
Also with packet size of 128 or 200 the problem persist.
This is big problem for VoIP !!!!!

Bye
Cuccio

changeip · Wed Oct 11, 2006 9:57 pm

What kind of hardware are you using with mikrotik? A 532 isn't powerful enough for that pps i dont think.

Sam

variable · Thu Oct 12, 2006 1:50 am

trying packing the tiny packets together with mikrotik packet packing protocol(m3p); i have yet to use this but it seems promising.

also i beleive nstreme does this by default, were you using nstreme or wds?

believewireless · Thu Oct 12, 2006 3:40 am

I just tested and we see the same thing. One link is NStreme2 and the other is NStreme w/WDS. Both show horrible performance with small UDP packets. TCP packets of the same size fly through at near full speed.

variable · Thu Oct 12, 2006 7:21 am

huh, anyone tried with m3p? does it max the cpu? as in, is it a cpu problem or radio/protocol limitation?

believewireless · Thu Oct 12, 2006 7:31 am

I don't see how M3P is going to help when transmitting 64byte packets works fine with TCP but not with UDP. M3P only mentions packet size and not UDP or TCP.

variable · Thu Oct 12, 2006 9:01 am

ya, i dont know, never used it.

does it appear to be a cpu limitation then?

Thu Oct 12, 2006 10:46 am

we have tested 13000-23000 pps on Rb532, with/without connection tracking respectively. m3p will not improve situation. check you testing methods, configuration, setup and connections.

believewireless · Thu Oct 12, 2006 3:11 pm

Have you tested with both UDP and TCP? Our setup is as follows:

1 GHz MT Router -> RB532 w/WDS -> RB532 w/WDS -> 1 GHz MT Router

We run the test between the 1 GHz routers using the Mikrotik bandwidth test.

/tool bandwidth-test XX.XX.XX.XX protocol=tcp local-tx-size=64 remote-tx-size=64 direction=transmit random-data=yes user=admin duration=5s

status: done testing
duration: 5s
tx-current: 15.6Mbps
tx-10-second-average: 16.4Mbps
tx-total-average: 16.4Mbps
random-data: yes
direction: transmit

Exact same test just changing to UDP:

tool bandwidth-test XX.XX.XX.XX protocol=udp local-tx-size=64 remote-tx-size=64 direction=transmit random-data=yes user=admin duration=5s

status: done testing
duration: 6s
tx-current: 2.0Mbps
tx-10-second-average: 1416.3kbps
tx-total-average: 1416.3kbps
random-data: yes
direction: transmit
tx-size: 64

RB532s and one 1GHz router are running 2.9.31 and the other 1 GHz router is 2.9.30.

cuccio · Thu Oct 12, 2006 4:46 pm

I have testing RB532 @330Mhz in 2 conditions:

1- one RB532 with Ether1 and Ether2 Bridged, No traking conn, PC AMD3000 in Ether1 and PCPentium IV 2800 in ETher2.

2- two RB532 with CM9 cards in Bridge, No traking conn, WDS and Nstreme (20Mhz channel) and the same computers.

I have testing also Supermicro server (PIV3200) with Ether 1000, max results 20.000pps

!!

The result don't change, poor performance.
The CPU of ALvarion is Motorola 266Mhz with 64Mb RAM

Any idea ??

Cuccio

cuccio · Thu Oct 12, 2006 4:47 pm

we have tested 13000-23000 pps on Rb532, with/without connection tracking respectively. m3p will not improve situation. check you testing methods, configuration, setup and connections.

Testing with 64Byte UDP packets ?????????
Sure ??????????

Cuccio

cuccio · Thu Oct 12, 2006 4:53 pm

Exact same test just changing to UDP:

tool bandwidth-test XX.XX.XX.XX protocol=udp local-tx-size=64 remote-tx-size=64 direction=transmit random-data=yes user=admin duration=5s

status: done testing
duration: 6s
tx-current: 2.0Mbps
tx-10-second-average: 1416.3kbps
tx-total-average: 1416.3kbps
random-data: yes
direction: transmit
tx-size: 64

RB532s and one 1GHz router are running 2.9.31 and the other 1 GHz router is 2.9.30.

Yesssss !!!!
Confirm, max 5Mbps in UDP with small pakets.

tpsretard · Thu Oct 12, 2006 6:10 pm

i just did this test also.

with UDP i could only get 1.06 mbits
with TCP i could get 13 mbits

end to end both rb532 single Nstreem

cuccio · Thu Oct 12, 2006 8:00 pm

Alvarion have a concatenation and bursting for small packets:

"Packet concatenation for improved VOIP latency"

RouterOS have bursting, but not concatenation.

Cuccio

jp · Fri Oct 13, 2006 5:35 am

I thought "framer policy" in nstreme was something like Alvarion's packet concatenation. If not, how does it differ?

We do mostly Alvarion links with MT on the ethernet side.

Fri Oct 13, 2006 11:29 am

tool bandwidth-test XX.XX.XX.XX protocol=udp local-tx-size=64 remote-tx-size=64 direction=transmit random-data=yes user=admin duration=5s

this command is not correct for testing.

you have set too small testing time, routeros hasn't even started the test yet, when you already interrupt it. set the time to few minutes

cuccio · Fri Oct 13, 2006 12:40 pm

After 6 minutes, the througput is 6Mbps !!

Cuccio

cuccio · Fri Oct 13, 2006 3:16 pm

Very very good program and GUI for traffic generation and analisys.
Good also for VoIP, generating VoIP packet with appropriate codec.

SW: http://www.grid.unina.it/software/ITG/
GUI: http://www.semken.com/projekte/index.html

Anyone have solution for UDP small packets ??

Bye Cuccio

tpsretard · Fri Oct 13, 2006 5:25 pm

after 5min still 1meg this is a problem

rmdjapri · Sat Oct 14, 2006 5:52 pm

So, no solution for this situation from MT ? I hope MT will do some checking about this thread. Online gaming using the small packets too.

sten · Thu Oct 19, 2006 12:15 am

This issue with small packets surprises you? LoL, did you think it was fiberoptics?

As a side note, 2x RB532 in lab running nstreme exact size 2560 with 24 mbit modulation and turbo gave me a solid 10 mbit at 100 byte sized packets. and that was -while- the rb's acted as bandwidth testers (which the cpus are the limit)! At 500 byte packets i got 35 mbit solid (still while rb's were bandwidth testers).

Lab is not equivalent to the real world unfortunately but i can assure you;
Thou dost naught ken that of which ye speake.

sten · Thu Oct 19, 2006 12:26 am

This issue with small packets surprises you? LoL, did you think it was fiberoptics?

As a side note, 2x RB532 in lab running nstreme exact size 2560 with 24 mbit modulation and turbo gave me a solid 10 mbit at 100 byte sized packets. and that was -while- the rb's acted as bandwidth testers (which the cpus are the limit)! At 500 byte packets i got 35 mbit solid (still while rb's were bandwidth testers).

Lab is not equivalent to the real world unfortunately but i can assure you;
Thou dost naught ken that of which ye speake.

karyal · Sat Oct 21, 2006 6:09 pm

This issue with small packets surprises you? LoL, did you think it was fiberoptics?

Uhm, well, i never thought to wireless as a fiber optics substitution, else no one would spend so much money to lay fibers..
I think we all know that one thing is lab test, one thing is real world..
But cuccio point is not at all wrong.. he doesn't understand why he gets poor performance on the link when using MT and not with an Alvarion stuff (and i do not see either, i have yet to see an Alvarion device perform better than a MT one, not to mention the costs).

It's not a matter of wireless or not, is more a matter of understanding if it's some kind of MT limit or configuration/situation/hardware..

Mon Oct 30, 2006 8:46 am

stuff is getting deleted for a reason. one of the possible reasons is when people post complete nonsense and are confusing people, which then complain about things that aren't true

jp · Mon Oct 30, 2006 7:25 pm

This issue with small packets surprises you? LoL, did you think it was fiberoptics?
But cuccio point is not at all wrong.. he doesn't understand why he gets poor performance on the link when using MT and not with an Alvarion stuff (and i do not see either, i have yet to see an Alvarion device perform better than a MT one, not to mention the costs).

I get 1ms latency on a 54mbps connection with Alvarion VL gear. It has packet concatenation (like nstreme framer policy), 10/20mhz channels, full snmp programming and monitoring, leds for alignment, solid firmware, and of course higher cost. Aside from the cost, it is very service provider friendly (easy to setup and maintain). It's really sort of understated.

believewireless · Thu Nov 02, 2006 9:04 pm

This is why I don't understand the issues we are seeing with Mikrotik. Our Alvarion VL gear gets 1-2ms to every CPE on the network regardless of the number of connected client radios.

With Mikrotik, we see horrible ping times using WDS and this UDP packet problem. I don't know if Mikrotik has a poor polling mechanism or what is causing the UDP problems.

I will say that the RouterOS3 looks like it is addressing some of the issues and I'd love to test it and see if it fixes the problems we are seeing.

I'm not expecting a "fiber" connection, but I would like it to perform like our Alvarion, Canopy, Trango, etc, etc. gear.

poor performance whit small packets !

poor performance whit small packets !

Who is online