1.1.2.1. Batching

The Batch QoS Policy can be used to decrease the amount of communication overhead associated with the transmission and (in the case of reliable communication) acknowledgment of small DDS samples, in order to increase throughput. It does this by collecting multiple user data DDS samples to be sent in a single network packet, to take advantage of the efficiency of sending larger packets.

Batching increases effective throughput dramatically for small data DDS samples. Throughput for small DDS samples is typically limited by CPU capacity and not by network bandwidth. Batching many smaller DDS samples to be sent in a single large packet increases network utilization and thus throughput in terms of DDS samples per second.

For more information, see BATCH QosPolicy, in the RTI Connext DDS Core Libraries User’s Manual.

The tests below show how powerful the use of batching can be. In these tests, we compare the RTI Perftest default configuration for throughput, which enables batching (with a batch size of 8192 bytes), and a configuration where batching is disabled.

Note

We choose 8192 bytes for the batch size because it is the closest power of two below the maximum MTU that can be configured for a regular NIC, which is 9000 bytes. This way the RTPS packet doesn’t need to be fragmented.

This configuration, by itself, cannot ensure that the packet will not be fragmented, since if the MTU of the NIC is set to a lower value (like the default value in many NICs, of 1500 bytes), then the packet will have fragmentation.

Note

Since some samples are delayed until the batch is ready, you may observe increased latency. The maximum delay, as well as maximum bytes or maximum number of samples, can be configured. See BATCH QosPolicy, in the RTI Connext DDS Core Libraries User’s Manual.

Unkeyed, UDPv4 10Gbps Network, C++98

Linux x64

Maximum Throughput

The graph below shows the expected throughput behavior when performing a 1-1 communication between two Linux nodes in a 10Gbps network with and without batching (with a batch size of 8192 bytes). The numbers are for best-effort as well as strict reliable reliability scenarios.

Detailed Statistics

This table contains the raw numbers presented by RTI Perftest. These numbers are the exact output with no further processing.

• Best Effort, Batching

Sample Size (Bytes)	Total Samples	Avg Samples/s	Avg Mbps	Lost Samples	Lost Samples (%)
32	98447835	4916862	1258.7	608768	0.61
64	91799845	4585163	2347.6	57216	0.06
128	80906785	4038802	4135.7	0	0.00
256	66036513	3298344	6755.0	19936	0.03
512	46765762	2336180	9569.0	0	0.00
1024	23812202	1190253	9750.6	0	0.00
2048	11949288	597398	9787.8	0	0.00
4096	4876518	243815	9889.3	1109040	18.53
8192	3005854	150283	9849.0	0	0.00
16384	1509117	75450	9889.5	79	0.01
32768	756061	37802	9909.8	16	0.00
63000	393526	19676	9916.8	8	0.00

• Best Effort, No Batching

Sample Size (Bytes)	Total Samples	Avg Samples/s	Avg Mbps	Lost Samples	Lost Samples (%)
32	7512988	375290	96.1	0	0.00
64	7501457	374714	191.9	0	0.00
128	7497866	374534	383.5	0	0.00
256	7154053	357364	781.9	0	0.00
512	7133771	356350	1559.6	0	0.00
1024	7044914	351913	3082.9	0	0.00
2048	6888612	344104	5937.8	0	0.00
4096	5927297	296076	9701.8	0	0.00
8192	3005854	150283	9849.0	0	0.00
16384	1509117	75450	9889.5	79	0.01
32768	756061	37802	9909.8	16	0.00
63000	393526	19676	9916.8	8	0.00

• Reliable, Batching

Sample Size (Bytes)	Total Samples	Avg Samples/s	Avg Mbps	Lost Samples	Lost Samples (%)
32	93771579	4686195	1199.7	0	0.00
64	84316173	4209912	2155.5	0	0.00
128	71793365	3584812	3670.8	0	0.00
256	54521333	2722685	5576.1	0	0.00
512	38241843	1909759	7822.4	0	0.00
1024	22874688	1142600	9360.2	0	0.00
2048	11953124	597062	9782.3	0	0.00
4096	5988600	299098	9800.9	0	0.00
8192	3004950	150226	9845.3	0	0.00
16384	1508850	75439	9888.0	0	0.00
32768	756049	37799	9908.9	0	0.00
63000	393534	19675	9916.4	0	0.00
100000	247815	12389	9911.5	0	0.00
500000	47611	2380	9523.5	0	0.00
1048576	16285	1012	9440.4	0	0.00
1548576	15096	755	9356.2	0	0.00
4194304	5598	280	9395.4	0	0.00
10485760	2055	103	8695.8	0	0.00

• Reliable, No Batching

Sample Size (Bytes)	Total Samples	Avg Samples/s	Avg Mbps	Lost Samples	Lost Samples (%)
32	4989963	249154	63.8	0	0.00
64	4938398	246654	126.3	0	0.00
128	4863735	242922	248.8	0	0.00
256	4892469	244359	500.4	0	0.00
512	4894170	244442	1001.2	0	0.00
1024	4876949	243583	1995.4	0	0.00
2048	4703617	234928	3849.1	0	0.00
4096	4450197	222272	7283.4	0	0.00
8192	3004950	150226	9845.3	0	0.00
16384	1508850	75439	9888.0	0	0.00
32768	756049	37799	9908.9	0	0.00
63000	393534	19675	9916.4	0	0.00

Perftest Scripts

To produce these tests, we executed RTI Perftest for C++98. The exact commands used can be found here:

Publisher Side

sudo /set_thr_mode.sh

echo EXECUTABLE IS $1
export executable=$1

echo OUTPUT PATH IS $2
export output_folder=$2

export exec_time=30
export nic=172.16.0.1
export pub_string="-pub \
        -transport UDPv4 \
        -nic $nic \
        -noPrint \
        -batchSize 0\
        -noOutputHeaders \
        -exec $exec_time \
        -noXML"

mkdir -p $output_folder

echo ">> UNKEYED BE N Batching"
export my_file=$output_folder/thr_udpv4_pub_unkeyed_be_noBatch.csv
touch $my_file
export extra_args=""
for index in 1 2 3; do
    for DATALEN in 32 64 128 256 512 1024 2048 4096 8192 16384 32768 63000; do
        export command="taskset -c 0 \
        $executable -best $pub_string -datalen $DATALEN $extra_args -batchSize 0"
        echo $command
        $command >> $my_file;
        sleep 3;
        export extra_args=" -noOutputHeaders "
    done
done
sleep 5;

echo ">> UNKEYED REL N Batching"
export my_file=$output_folder/thr_udpv4_pub_unkeyed_rel_noBatch.csv
touch $my_file
export extra_args=""
for index in 1 2 3; do
    for DATALEN in 32 64 128 256 512 1024 2048 4096 8192 16384 32768 63000; do
        export command="taskset -c 0 \
        $executable $pub_string -datalen $DATALEN $extra_args -batchSize 0"
        echo $command
        $command >> $my_file;
        sleep 3;
        export extra_args=" -noOutputHeaders "
    done
done
sleep 5;

Subscriber Side

sudo /set_thr_mode.sh

echo EXECUTABLE IS $1
export executable=$1

echo OUTPUT PATH IS $2
export output_folder=$2

export nic=172.16.0.2
export sub_string="-sub \
        -transport UDPv4 \
        -nic $nic \
        -noPrint \
        -noOutputHeaders \
        -noXML"

mkdir -p $output_folder

echo ">> UNKEYED BE"
export my_file=$output_folder/thr_udpv4_sub_unkeyed_be_noBatch.csv
touch $my_file
export extra_args=""
for index in 1 2 3; do
    for DATALEN in 32 64 128 256 512 1024 2048 4096 8192 16384 32768 63000; do
        export command="taskset -c 0 \
        $executable -best -datalen $DATALEN $sub_string $extra_args"
        echo $command ---- $index
        $command >> $my_file;
        sleep 10;
        export extra_args=" -noOutputHeaders "
    done
done
sleep 5;


echo ">> UNKEYED REL"
export my_file=$output_folder/thr_udpv4_sub_unkeyed_rel_noBatch.csv
touch $my_file
export extra_args=""
for index in 1 2 3; do
    for DATALEN in 32 64 128 256 512 1024 2048 4096 8192 16384 32768 63000; do
        export command="taskset -c 0 \
        $executable -datalen $DATALEN $sub_string $extra_args"
        echo $command ---- $index
        $command >> $my_file;
        sleep 10;
        export extra_args=" -noOutputHeaders "
    done
done
sleep 5;

Test Hardware

The following hardware was used to perform these tests:

Linux Nodes

Processor: Intel® Xeon® E-2186G 3.8GHz, 12M cache, 6C/12T, turbo (95W)
RAM: 16GB 2666MT/s DDR4 ECC UDIMM
NIC 1: Intel X550 Dual Port 10GbE BASE-T Adapter, PCIe Full Height
NIC 2: Intel Ethernet I350 Dual Port 1GbE BASE-T Adapter, PCIe Low Profile
OS: Ubuntu 18.04 -- gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0

Switch

Dell Networking S4048T-ON, 48x 10GBASE-T and 6x 40GbE QSFP+ ports, IO to PSU air, 2x AC PSU, OS9