UDP data rate slow with 1GBE ethernet connection

hi folks,

i have a two work station that are connects via 1GBE ethernet.

when transfering data at this setup i've encountered that the data rate is ~20MBs which is mach lower then the expected data rate of a 1GBe comminication. i've used udp object and udpport object that ended up with the same results in MATLAB. Code below>>

ferther investing this ive found out that an off the shelf Cpp code gets 800MBs data rate.

• is there a way to enhance or controll the data rate from within MATLAB?
• if not how can i work with higher data rates inder my setup, use external C\Java code?

--------------------------------------- Code ---------------------------------------

% TASK send data from port A to Port B

% UDP test for the same computer nIP = 1

% UDP test for the two computers nIP = 2

clc, clear all, close all

n_packets = 200;

packet_size = 4096;

BufferSize = packet_size * 2;

nIP = 1;

%

ipA = '192.168.1.100'; portA = 9092;

if nIP == 1

ipB = '192.168.1.100'; portB = 9093;

else

ipB = '192.168.1.101'; portB = 9093;

end

u=udp(ipB,'RemotePort',portB,'Localport',portA);

set(u, 'InputBufferSize', BufferSize);

set(u, 'OutputBufferSize', BufferSize);

set(u,'DatagramTerminateMode','off')

set(u,'Timeout',0.01);

a = zeros(packet_size,1,'int16');

b = zeros(packet_size,1,'int16');

a = ones(packet_size,1,'int16');

i = 1;

miss_count = 0;

fopen(u);

tic

% Loop over write

while 1 %i<n_packets

a(1) = mod(i,32767);

fwrite(u,a,'int16');

i = i + 1;

end

t_proc = toc;

echoudp('off');

fclose(u);

delete(u);

data_rate = ((n_packets * packet_size * 16)/1e6)/t_proc;

fprintf('\n n_miss_packet = %g',miss_count);

fprintf('\n tudp_trx_sec = %g',t_proc);

fprintf('\n udp_rate_Mbit/s = %g',data_rate);

% TASK get data from port A to Port B

% TASK send data from port A to Port B

% UDP test for the same computer nIP = 1

% UDP test for the two computers nIP = 2

clc, clear, close all

n_packets = 200;

packet_size = 4096; % 16 bits

BufferSize = packet_size * 2;

nIP = 1;

%Shimmi

ii_max = 15;

t_proc_sum = zeros(size(1:ii_max));

miss_count_sum = zeros(size(1:ii_max));

date_rate_sum = zeros(size(1:ii_max));

ipA = '10.1.21.81'; portA = 9092;

if nIP == 1

ipB = '192.168.1.100'; portB = 9093;

else

ipB = '192.168.1.100'; portB = 9093;

end

for ii=1:ii_max

u=udp(ipA,'RemotePort',portA,'Localport',portB);

set(u, 'InputBufferSize', BufferSize);

set(u, 'OutputBufferSize', BufferSize);

a = zeros(packet_size,1,'int16');

b = zeros(packet_size,1,'int16');

a = ones(packet_size,1,'int16');

set(u,'DatagramTerminateMode','off')

set(u,'Timeout',0.01);

i = 1;

iprev = 1;

miss_count = 0;

% Open UDP

fopen(u);

tic

% Loop over read

while i<n_packets

b=fread(u,packet_size,'int16');

icurr = b(1);

if (iprev ~= icurr)

miss_count = miss_count + 1;

end

iprev = icurr + 1;

i = i + 1;

end

t_proc = toc;

echoudp('off');

fclose(u);

delete(u);

data_rate = ((n_packets * packet_size * 16)/t_proc)/1e6;

%

% fprintf('\n n_miss_packet = %g',miss_count);

% fprintf('\n tudp_rx_sec = %g',t_proc);

% fprintf('\n udp_rate_Mbit/s = %g',data_rate);

%analytics

miss_count_sum(ii) = miss_count;

t_proc_sum (ii) =t_proc;

date_rate_sum (ii) = data_rate;

ii = ii+1;

end

fprintf('\n n_miss_packet = %f',mean(miss_count_sum));

fprintf('\n tudp_rx_sec = %f',mean(t_proc_sum));

fprintf('\n udp_rate_Mbit/s = %f',mean(date_rate_sum));

fprintf('\n ');

[Jeffrey Clark]

@Shimon Elkin, have you tried the preferred Connect to UDP socket - MATLAB (mathworks.com)?

[Shimon Elkin]

Hi @Jeffrey Clark, yes i've used udp object and udpport object that ended up with the same results in MATLAB.

Answer 1

[Vinayak]

Hi Shimon,

The issues with UDP performance that you've outlined could stem from various factors. To address this, I recommend the following actions:

• Confirm that the network infrastructure can accommodate the required transfer speeds.
• Optimize the Buffer Size and Packet Sizes; experimenting with different combinations of values can help find the optimal balance, minimizing overhead and fragmentation.
• Explore the potential of parallel processing capabilities, if applicable to your specific problem.
• Consider utilizing external language interfaces such as C or Python.

For further information on “udpport”, and the Parallel Computing Toolbox, please refer to the following links:

• UDP Port: https://www.mathworks.com/help/instrument/udpport.html
• External Interfaces: https://www.mathworks.com/help/matlab/external-language-interfaces.html
• Parallel Computing Toolbox: https://www.mathworks.com/help/parallel-computing/index.html

Hope this helps!