2.2. Performance Guide

2.2.1. Kernel Performance Guide

Read This First

All performance numbers provided in this document are gathered using following Evaluation Modules unless otherwise specified.

Name Description
J7200 EVM J7200 Evaluation Module rev E1 with ARM running at 2GHz, DDR data rate 2666 MT/S, L3 Cache size 3MB

Table: Evaluation Modules


About This Manual

This document provides performance data for each of the device drivers which are part of the Process SDK Linux package. This document should be used in conjunction with release notes and user guides provided with the Process SDK Linux package for information on specific issues present with drivers included in a particular release.

If You Need Assistance

For further information or to report any problems, contact http://e2e.ti.com/ or http://support.ti.com/

2.2.1.1. System Benchmarks

2.2.1.1.1. LMBench

LMBench is a collection of microbenchmarks of which the memory bandwidth and latency related ones are typically used to estimate processor memory system performance. More information about lmbench at http://lmbench.sourceforge.net/whatis_lmbench.html and http://lmbench.sourceforge.net/man/lmbench.8.html

Latency: lat_mem_rd-stride128-szN, where N is equal to or smaller than the cache size at given level measures the cache miss penalty. N that is at least double the size of last level cache is the latency to external memory.

Bandwidth: bw_mem_bcopy-N, where N is equal to or smaller than the cache size at a given level measures the achievable memory bandwidth from software doing a memcpy() type operation. Typical use is for external memory bandwidth calculation. The bandwidth is calculated as byte read and written counts as 1 which should be roughly half of STREAM copy result.

Benchmarks j7200-evm: perf
af_unix_sock_stream_latency (microsec) 19.75
af_unix_socket_stream_bandwidth (MBs) 1500.43
bw_file_rd-io-1mb (MB/s) 1626.35
bw_file_rd-o2c-1mb (MB/s) 916.09
bw_mem-bcopy-16mb (MB/s) 1518.31
bw_mem-bcopy-1mb (MB/s) 1461.45
bw_mem-bcopy-2mb (MB/s) 1435.24
bw_mem-bcopy-4mb (MB/s) 1484.23
bw_mem-bcopy-8mb (MB/s) 1515.15
bw_mem-bzero-16mb (MB/s) 1141.63
bw_mem-bzero-1mb (MB/s) 1555.13 (min 1461.45, max 1648.80)
bw_mem-bzero-2mb (MB/s) 1308.55 (min 1181.86, max 1435.24)
bw_mem-bzero-4mb (MB/s) 1312.32 (min 1140.41, max 1484.23)
bw_mem-bzero-8mb (MB/s) 1328.27 (min 1141.39, max 1515.15)
bw_mem-cp-16mb (MB/s) 965.43
bw_mem-cp-1mb (MB/s) 1261.22 (min 947.33, max 1575.11)
bw_mem-cp-2mb (MB/s) 1064.59 (min 936.04, max 1193.14)
bw_mem-cp-4mb (MB/s) 1041.38 (min 944.29, max 1138.47)
bw_mem-cp-8mb (MB/s) 1046.64 (min 953.18, max 1140.09)
bw_mem-fcp-16mb (MB/s) 1526.86
bw_mem-fcp-1mb (MB/s) 1554.46 (min 1460.12, max 1648.80)
bw_mem-fcp-2mb (MB/s) 1312.17 (min 1181.86, max 1442.48)
bw_mem-fcp-4mb (MB/s) 1321.33 (min 1140.41, max 1502.25)
bw_mem-fcp-8mb (MB/s) 1322.44 (min 1141.39, max 1503.48)
bw_mem-frd-16mb (MB/s) 5592.45
bw_mem-frd-1mb (MB/s) 3912.46 (min 1460.12, max 6364.79)
bw_mem-frd-2mb (MB/s) 3501.59 (min 1442.48, max 5560.70)
bw_mem-frd-4mb (MB/s) 3524.09 (min 1502.25, max 5545.93)
bw_mem-frd-8mb (MB/s) 3535.32 (min 1503.48, max 5567.15)
bw_mem-fwr-16mb (MB/s) 1140.33
bw_mem-fwr-1mb (MB/s) 3969.95 (min 1575.11, max 6364.79)
bw_mem-fwr-2mb (MB/s) 3376.92 (min 1193.14, max 5560.70)
bw_mem-fwr-4mb (MB/s) 3342.20 (min 1138.47, max 5545.93)
bw_mem-fwr-8mb (MB/s) 3353.62 (min 1140.09, max 5567.15)
bw_mem-rd-16mb (MB/s) 5085.02
bw_mem-rd-1mb (MB/s) 3633.30 (min 994.20, max 6272.40)
bw_mem-rd-2mb (MB/s) 2964.12 (min 765.99, max 5162.24)
bw_mem-rd-4mb (MB/s) 2903.09 (min 746.55, max 5059.63)
bw_mem-rd-8mb (MB/s) 2895.00 (min 744.26, max 5045.73)
bw_mem-rdwr-16mb (MB/s) 745.30
bw_mem-rdwr-1mb (MB/s) 967.82 (min 947.33, max 988.31)
bw_mem-rdwr-2mb (MB/s) 852.05 (min 768.05, max 936.04)
bw_mem-rdwr-4mb (MB/s) 846.05 (min 747.80, max 944.29)
bw_mem-rdwr-8mb (MB/s) 852.39 (min 751.60, max 953.18)
bw_mem-wr-16mb (MB/s) 743.32
bw_mem-wr-1mb (MB/s) 991.26 (min 988.31, max 994.20)
bw_mem-wr-2mb (MB/s) 767.02 (min 765.99, max 768.05)
bw_mem-wr-4mb (MB/s) 747.18 (min 746.55, max 747.80)
bw_mem-wr-8mb (MB/s) 747.93 (min 744.26, max 751.60)
bw_mmap_rd-mo-1mb (MB/s) 6305.17
bw_mmap_rd-o2c-1mb (MB/s) 1177.06
bw_pipe (MB/s) 599.84
bw_unix (MB/s) 1500.43
lat_connect (us) 47.99
lat_ctx-2-128k (us) 4.44
lat_ctx-2-256k (us) 12.20
lat_ctx-4-128k (us) 4.04
lat_ctx-4-256k (us) 6.33
lat_fs-0k (num_files) 516.00
lat_fs-10k (num_files) 133.00
lat_fs-1k (num_files) 241.00
lat_fs-4k (num_files) 290.00
lat_mem_rd-stride128-sz1000k (ns) 17.46
lat_mem_rd-stride128-sz125k (ns) 5.17
lat_mem_rd-stride128-sz250k (ns) 5.18
lat_mem_rd-stride128-sz31k (ns) 3.20
lat_mem_rd-stride128-sz50 (ns) 2.00
lat_mem_rd-stride128-sz500k (ns) 12.12
lat_mem_rd-stride128-sz62k (ns) 5.16
lat_mmap-1m (us) 28.00
lat_ops-double-add (ns) 0.32
lat_ops-double-mul (ns) 2.00
lat_ops-float-add (ns) 0.32
lat_ops-float-mul (ns) 2.00
lat_ops-int-add (ns) 0.50
lat_ops-int-bit (ns) 0.33
lat_ops-int-div (ns) 4.01
lat_ops-int-mod (ns) 4.68
lat_ops-int-mul (ns) 1.52
lat_ops-int64-add (ns) 0.50
lat_ops-int64-bit (ns) 0.33
lat_ops-int64-div (ns) 3.00
lat_ops-int64-mod (ns) 5.68
lat_pagefault (us) 0.59
lat_pipe (us) 11.23
lat_proc-exec (us) 762.43
lat_proc-fork (us) 655.00
lat_proc-proccall (us) 0.00
lat_select (us) 14.33
lat_sem (us) 1.37
lat_sig-catch (us) 2.59
lat_sig-install (us) 0.49
lat_sig-prot (us) 0.30
lat_syscall-fstat (us) 0.67
lat_syscall-null (us) 0.35
lat_syscall-open (us) 206.36
lat_syscall-read (us) 0.42
lat_syscall-stat (us) 1.50
lat_syscall-write (us) 0.44
lat_tcp (us) 0.72
lat_unix (us) 19.75
latency_for_0.50_mb_block_size (nanosec) 12.12
latency_for_1.00_mb_block_size (nanosec) 8.73 (min 0.00, max 17.46)
pipe_bandwidth (MBs) 599.84
pipe_latency (microsec) 11.23
procedure_call (microsec) 0.00
select_on_200_tcp_fds (microsec) 14.33
semaphore_latency (microsec) 1.37
signal_handler_latency (microsec) 0.49
signal_handler_overhead (microsec) 2.59
tcp_ip_connection_cost_to_localhost (microsec) 47.99
tcp_latency_using_localhost (microsec) 0.72

Table: LM Bench Metrics

2.2.1.1.2. Dhrystone

Dhrystone is a core only benchmark that runs from warm L1 caches in all modern processors. It scales linearly with clock speed. For standard ARM cores the DMIPS/MHz score will be identical with the same compiler and flags.

Execute the benchmark with the following:

runDhrystone
Benchmarks j7200-evm: perf
cpu_clock (MHz) 2000.00
dhrystone_per_mhz (DMIPS/MHz) 3.60
dhrystone_per_second (DhrystoneP) 12500000.00

Table: Dhrystone Benchmark

2.2.1.1.3. Whetstone

Whetstone is a benchmark primarily measuring floating-point arithmetic performance.

Execute the benchmark with the following:

runWhetstone
Benchmarks j7200-evm: perf
whetstone (MIPS) 10000.00

Table: Whetstone Benchmark

2.2.1.1.4. Linpack

Linpack measures peak double precision (64 bit) floating point performance in solving a dense linear system.

Benchmarks j7200-evm: perf
linpack (Kflops) 2507913.00

Table: Linpack Benchmark

2.2.1.1.5. NBench

NBench which stands for Native Benchmark is used to measure macro benchmarks for commonly used operations such as sorting and analysis algorithms. More information about NBench at https://en.wikipedia.org/wiki/NBench and https://nbench.io/articles/index.html

Benchmarks j7200-evm: perf
assignment (Iterations) 29.65
fourier (Iterations) 54417.00
fp_emulation (Iterations) 249.98
huffman (Iterations) 2426.90
idea (Iterations) 7996.50
lu_decomposition (Iterations) 1413.80
neural_net (Iterations) 26.38
numeric_sort (Iterations) 877.94
string_sort (Iterations) 424.02

Table: NBench Benchmarks

2.2.1.1.6. CoreMarkPro

CoreMark®-Pro is a comprehensive, advanced processor benchmark that works with and enhances the market-proven industry-standard EEMBC CoreMark® benchmark. While CoreMark stresses the CPU pipeline, CoreMark-Pro tests the entire processor, adding comprehensive support for multicore technology, a combination of integer and floating-point workloads, and data sets for utilizing larger memory subsystems.

Benchmarks j7200-evm: perf
cjpeg-rose7-preset (workloads/) 82.64
core (workloads/) 0.78
coremark-pro () 2129.31
linear_alg-mid-100x100-sp (workloads/) 80.78
loops-all-mid-10k-sp (workloads/) 2.23
nnet_test (workloads/) 3.60
parser-125k (workloads/) 9.43
radix2-big-64k (workloads/) 93.66
sha-test (workloads/) 158.73
zip-test (workloads/) 43.48

Table: CoreMarkPro

Benchmarks j7200-evm: perf
cjpeg-rose7-preset (workloads/) 161.29
core (workloads/) 1.55
coremark-pro () 3260.79
linear_alg-mid-100x100-sp (workloads/) 158.73
loops-all-mid-10k-sp (workloads/) 3.06
nnet_test (workloads/) 7.14
parser-125k (workloads/) 7.49
radix2-big-64k (workloads/) 75.61
sha-test (workloads/) 312.50
zip-test (workloads/) 76.92

Table: CoreMarkPro for Two Cores

2.2.1.1.7. MultiBench

MultiBench™ is a suite of benchmarks that allows processor and system designers to analyze, test, and improve multicore processors. It uses three forms of concurrency: Data decomposition: multiple threads cooperating on achieving a unified goal and demonstrating a processor’s support for fine grain parallelism. Processing multiple data streams: uses common code running over multiple threads and demonstrating how well a processor scales over scalable data inputs. Multiple workload processing: shows the scalability of general-purpose processing, demonstrating concurrency over both code and data. MultiBench combines a wide variety of application-specific workloads with the EEMBC Multi-Instance-Test Harness (MITH), compatible and portable with most any multicore processors and operating systems. MITH uses a thread-based API (POSIX-compliant) to establish a common programming model that communicates with the benchmark through an abstraction layer and provides a flexible interface to allow a wide variety of thread-enabled workloads to be tested.

Benchmarks j7200-evm: perf
4m-check (workloads/) 787.65
4m-check-reassembly (workloads/) 105.93
4m-check-reassembly-tcp (workloads/) 80.39
4m-check-reassembly-tcp-cmykw2-rotatew2 (workloads/) 33.41
4m-check-reassembly-tcp-x264w2 (workloads/) 2.56
4m-cmykw2 (workloads/) 302.12
4m-cmykw2-rotatew2 (workloads/) 50.21
4m-reassembly (workloads/) 93.11
4m-rotatew2 (workloads/) 55.49
4m-tcp-mixed (workloads/) 258.07
4m-x264w2 (workloads/) 2.62
idct-4m (workloads/) 34.70
idct-4mw1 (workloads/) 34.71
ippktcheck-4m (workloads/) 782.96
ippktcheck-4mw1 (workloads/) 784.44
ipres-4m (workloads/) 143.13
ipres-4mw1 (workloads/) 140.71
md5-4m (workloads/) 39.54
md5-4mw1 (workloads/) 40.39
rgbcmyk-4m (workloads/) 161.94
rgbcmyk-4mw1 (workloads/) 162.08
rotate-4ms1 (workloads/) 50.56
rotate-4ms1w1 (workloads/) 51.02
rotate-4ms64 (workloads/) 51.60
rotate-4ms64w1 (workloads/) 50.81
x264-4mq (workloads/) 1.41
x264-4mqw1 (workloads/) 1.40

Table: Multibench

2.2.1.2. Boot-time Measurement

2.2.1.2.1. Boot media: MMCSD

Boot Configuration j7200-evm: boot time (sec)
Kernel boot time test when bootloader, kernel and sdk-rootfs are in mmc-sd 17.28 (min 17.16, max 17.46)
Kernel boot time test when init is /bin/sh and bootloader, kernel and sdk-rootfs are in mmc-sd 4.87 (min 4.83, max 4.88)

Table: Boot time MMC/SD


2.2.1.3. Ethernet

Ethernet performance benchmarks were measured using Netperf 2.7.1 https://hewlettpackard.github.io/netperf/doc/netperf.html Test procedures were modeled after those defined in RFC-2544: https://tools.ietf.org/html/rfc2544, where the DUT is the TI device and the “tester” used was a Linux PC. To produce consistent results, it is recommended to carry out performance tests in a private network and to avoid running NFS on the same interface used in the test. In these results, CPU utilization was captured as the total percentage used across all cores on the device, while running the performance test over one external interface.

UDP Throughput (0% loss) was measured by the procedure defined in RFC-2544 section 26.1: Throughput. In this scenario, netperf options burst_size (-b) and wait_time (-w) are used to limit bandwidth during different trials of the test, with the goal of finding the highest rate at which no loss is seen. For example, to limit bandwidth to 500Mbits/sec with 1472B datagram:

burst_size = <bandwidth (bits/sec)> / 8 (bits -> bytes) / <UDP datagram size> / 100 (seconds -> 10 ms)
burst_size = 500000000 / 8 / 1472 / 100 = 425

wait_time = 10 milliseconds (minimum supported by Linux PC used for testing)

UDP Throughput (possible loss) was measured by capturing throughput and packet loss statistics when running the netperf test with no bandwidth limit (remove -b/-w options).

In order to start a netperf client on one device, the other device must have netserver running. To start netserver:

netserver [-p <port_number>] [-4 (IPv4 addressing)] [-6 (IPv6 addressing)]

Running the following shell script from the DUT will trigger netperf clients to measure bidirectional TCP performance for 60 seconds and report CPU utilization. Parameter -k is used in client commands to summarize selected statistics on their own line and -j is used to gain additional timing measurements during the test.

#!/bin/bash
for i in 1
do
   netperf -H <tester ip> -j -c -l 60 -t TCP_STREAM --
      -k DIRECTION,THROUGHPUT,MEAN_LATENCY,LOCAL_CPU_UTIL,REMOTE_CPU_UTIL,LOCAL_BYTES_SENT,REMOTE_BYTES_RECVD,LOCAL_SEND_SIZE &

   netperf -H <tester ip> -j -c -l 60 -t TCP_MAERTS --
      -k DIRECTION,THROUGHPUT,MEAN_LATENCY,LOCAL_CPU_UTIL,REMOTE_CPU_UTIL,LOCAL_BYTES_SENT,REMOTE_BYTES_RECVD,LOCAL_SEND_SIZE &
done

Running the following commands will trigger netperf clients to measure UDP burst performance for 60 seconds at various burst/datagram sizes and report CPU utilization.

  • For UDP egress tests, run netperf client from DUT and start netserver on tester.
netperf -H <tester ip> -j -c -l 60 -t UDP_STREAM -b <burst_size> -w <wait_time> -- -m <UDP datagram size>
   -k DIRECTION,THROUGHPUT,MEAN_LATENCY,LOCAL_CPU_UTIL,REMOTE_CPU_UTIL,LOCAL_BYTES_SENT,REMOTE_BYTES_RECVD,LOCAL_SEND_SIZE
  • For UDP ingress tests, run netperf client from tester and start netserver on DUT.
netperf -H <DUT ip> -j -C -l 60 -t UDP_STREAM -b <burst_size> -w <wait_time> -- -m <UDP datagram size>
   -k DIRECTION,THROUGHPUT,MEAN_LATENCY,LOCAL_CPU_UTIL,REMOTE_CPU_UTIL,LOCAL_BYTES_SENT,REMOTE_BYTES_RECVD,LOCAL_SEND_SIZE

2.2.1.3.1. CPSW/CPSW2g/CPSW3g Ethernet Driver

  • CPSW2g: AM65x, J7200, J721e, J721S2, J784S4
  • CPSW3g: AM64x

TCP Bidirectional Throughput

Command Used j7200-evm: THROUGHPUT (Mbits/sec) j7200-evm: CPU Load % (LOCAL_CPU_UTIL)
netperf -H 192.168.0.1 -j -c -C -l 60 -t TCP_STREAM; netperf -H 192.168.0.1 -j -c -C -l 60 -t TCP_MAERTS 1407.98 61.25

Table: CPSW TCP Bidirectional Throughput |

UDP Throughput

Frame Size(bytes) j7200-evm: UDP Datagram Size(bytes) (LOCAL_SEND_SIZE) j7200-evm: THROUGHPUT (Mbits/sec) j7200-evm: CPU Load % (LOCAL_CPU_UTIL)
64 18.00 32.06 92.89
256 210.00 387.25 92.62
512 466.00 873.69 92.67
1024 978.00 495.71 32.52
1518 1472.00 956.84 47.60

Table: CPSW UDP Egress Throughput |

Frame Size(bytes) j7200-evm: UDP Datagram Size(bytes) (LOCAL_SEND_SIZE) j7200-evm: THROUGHPUT (Mbits/sec) j7200-evm: CPU Load % (LOCAL_CPU_UTIL)
64 18.00 9.42 35.89
128 82.00 45.59 39.78
256 210.00 66.53 20.43
1024 978.00 228.46 19.46

Table: CPSW UDP Ingress Throughput (0% loss)


Frame Size(bytes) j7200-evm: UDP Datagram Size(bytes) (LOCAL_SEND_SIZE) j7200-evm: THROUGHPUT (Mbits/sec) j7200-evm: CPU Load % (LOCAL_CPU_UTIL) j7200-evm: Packet Loss %
64 18.00 22.90 73.00 0.03
128 82.00 96.88 71.85 8.20
256 210.00 374.43 80.52 3.91
1024 978.00 934.69 81.03 0.01

Table: CPSW UDP Ingress Throughput (possible loss)


2.2.1.4. PCIe Driver

2.2.1.4.1. PCIe-NVMe-SSD

2.2.1.4.1.1. J7200-EVM
Buffer size (bytes) j7200-evm: Write EXT4 Throughput (Mbytes/sec) j7200-evm: Write EXT4 CPU Load (%) j7200-evm: Read EXT4 Throughput (Mbytes/sec) j7200-evm: Read EXT4 CPU Load (%)
1m 720.00 20.07 1524.00 18.61
4m 716.00 16.32 1523.00 10.61
4k 81.70 50.79 124.00 47.52
256k 738.00 27.77 1517.00 34.79
  • Filesize used is: 10G
  • FIO command options: –ioengine=libaio –iodepth=4 –numjobs=1 –direct=1 –runtime=60 –time_based
  • Platform: Speed 8GT/s, Width x2
  • SSD being used: PLEXTOR PX-128M8PeY

2.2.1.5. OSPI Flash Driver

2.2.1.5.1. J7200-EVM

2.2.1.5.1.1. UBIFS
Buffer size (bytes) j7200-evm: Write UBIFS Throughput (Mbytes/sec) j7200-evm: Write UBIFS CPU Load (%) j7200-evm: Read UBIFS Throughput (Mbytes/sec) j7200-evm: Read UBIFS CPU Load (%)
102400 0.17 (min 0.12, max 0.26) 41.67 (min 40.19, max 42.32) 73.81 57.14
262144 0.14 (min 0.11, max 0.18) 41.75 (min 40.70, max 42.30) 71.74 40.00
524288 0.14 (min 0.11, max 0.18) 42.53 (min 42.17, max 42.99) 73.89 25.00
1048576 0.14 (min 0.11, max 0.18) 42.02 (min 41.28, max 42.76) 72.76 20.00
2.2.1.5.1.2. RAW
File size (Mbytes) j7200-evm: Raw Read Throughput (Mbytes/sec)
50 192.31

2.2.1.6. UBoot QSPI/OSPI Driver

2.2.1.6.1. J7200-EVM

File size (bytes in hex) j7200-evm: Write Throughput (Kbytes/sec) j7200-evm: Read Throughput (Kbytes/sec)
400000 394.42 195047.62
800000 398.48 240941.18
1000000 399.41 277694.92
2000000 404.26 300623.85

2.2.1.7. UBoot EMMC Driver


2.2.1.7.1. J7200-EVM


File size (bytes in hex) j7200-evm: Write Throughput (Kbytes/sec) j7200-evm: Read Throughput (Kbytes/sec)
2000000 59686.70 306242.99
4000000 60457.56 303407.41

2.2.1.8. MMC/SD Driver

Warning

IMPORTANT: The performance numbers can be severely affected if the media is mounted in sync mode. Hot plug scripts in the filesystem mount removable media in sync mode to ensure data integrity. For performance sensitive applications, umount the auto-mounted filesystem and re-mount in async mode.


2.2.1.8.1. J7200-EVM


Buffer size (bytes) j7200-evm: Write EXT4 Throughput (Mbytes/sec) j7200-evm: Write EXT4 CPU Load (%) j7200-evm: Read EXT4 Throughput (Mbytes/sec) j7200-evm: Read EXT4 CPU Load (%)
1m 19.10 1.11 85.70 1.62
4m 19.10 0.93 86.20 1.32
4k 4.84 5.20 16.50 11.11
256k 18.80 1.41 84.10 2.62


The performance numbers were captured using the following:

  • SanDisk 8GB MicroSDHC Class 10 Memory Card
  • Partition was mounted with async option

2.2.1.9. UBoot MMC/SD Driver


2.2.1.9.1. J7200-EVM

File size (bytes in hex) j7200-evm: Write Throughput (Kbytes/sec) j7200-evm: Read Throughput (Kbytes/sec)
400000 17579.40 78769.23
800000 17770.07 86231.58
1000000 32572.56 90021.98

The performance numbers were captured using the following:

  • SanDisk 8GB MicroSDHC Class 10 Memory Card

2.2.1.10. CRYPTO Driver

2.2.1.10.1. IPSec Software Performance

Algorithm j7200-evm: Throughput (Mbps) j7200-evm: Packets/Sec j7200-evm: CPU Load
3des 197.20 17.00 48.90
aes128 450.70 40.00 57.45
aes192 445.00 39.00 56.74
aes256 445.60 39.00 57.60

2.2.1.11. DCAN Driver

Performance and Benchmarks not available in this release.