How to limit rate?

Hi, 

Will check and get back to you soon. 

Best regards, 

Sudheer

Hi,

It doesn't work as expected. 1 and 2 is the test result of iperf3 using tcp. 3 and 4 is the test result of iperf3 using udp. The ale entry and policer are in the same configuration as before.

Is this a chip problem or a sdk problem?

Best Regards,

Lei

Hi,

I have tested at our end, It was working as expected in UDP.
As I have mentioned in TCP it was due to Handshake we are not getting the desired rate.

Are you running client on remote side i.e. PC side or not? If not, can you check by running client on PC side.
You have to run from PC side as policer is created on match from SA of PC, DA of SOC A72 client.

Best Regards,
Sudheer

Hi,

The global policer configuration is as follows. 

policerEn: 1, yellowDropEn: 1, redDropEn: 1, yellowThress: 0, policerNoMatchMode: 0, noMatchPolicer: -1558156804

hit 1 indicates that this policer setting has been hit via the log file. In this policer, it can be seen that some of the packets are colored yellow and red. According to our global configuration, all yellow and red colored packets should be discarded. cir is set to 10000000 (10M) and pir is set to 20000000 (20M). So the udp speed tested by iperf3 should be 10M to be reasonable, but the experimental results show that it is only about 6M. And the tcp speed tested via iperf3 is just illogical. 

[MCU2_0]  18774.194348 s: src/mod/cpsw_ale.c line[5624]:     0: POLICER_PORT, PORT_NUM: 3 ISTRUNK:0
[MCU2_0]  18774.194388 s: src/mod/cpsw_ale.c line[5642]:     0: POLICER_DST_MAC,ALE Index: 11
[MCU2_0]  18774.194424 s: src/mod/cpsw_ale.c line[5648]:     0: POLICER_SRC_MAC,ALE Index: 10
[MCU2_0]  18774.194460 s: src/mod/cpsw_ale.c line[5672]:     0: POLICER_SRC_IP,ALE Index: 12
[MCU2_0]  18774.194496 s: src/mod/cpsw_ale.c line[5678]:     0: POLICER_DST_IP,ALE Index: 13
[MCU2_0]  18774.194531 s: src/mod/cpsw_ale.c line[5684]:     0: POLICER_THREAD,THREAD ID:0
[MCU2_0]  18774.194566 s: src/mod/cpsw_ale.c line[5692]:     0: POLICER_PIR_IDL_INC_VAL: 2048
[MCU2_0]  18774.194602 s: src/mod/cpsw_ale.c line[5701]:     0: POLICER_CIR_IDL_INC_VAL: 1024
[MCU2_0]  18774.194646 s: src/mod/cpsw_ale.c line[5729]:     0: POLICER_STATS: Hit: 1, RedHit: 1, YellowHit: 1

Does this mean there is no way to speed limit tcp via the chip?

Doredla Sudheer Kumar said:

As I have mentioned in TCP it was due to Handshake we are not getting the desired rate.

Finally I would like to ask if this is a software problem or a chip problem? If it's a software problem, is it possible to fix the problem in sdk0703?

Best Regards,
Lei

Hi,

When we tested it before, tda4 A72 was the receiver (iperf3 -s) and another ecu was the sender (iperf3 -c 192.168.4.153 -b 40M -t 10s). After I reversed them today, the test results did not change.

I have two questions about the speed limit.

First, why is it that I set pir to 20Mb,cir to 10Mb, but the test result via iperf3 about udp is only about 6Mb. Isn't the difference too big?

Secondly, why is it that as long as pir and cir are set, the tcp sender's speed is very low, but as long as pir and cir are not set, the tested speed about tcp is the speed set by iperf3.

Best Regards,
Lei

Hi,

Lei Zhang said:

When we tested it before, tda4 A72 was the receiver (iperf3 -s) and another ecu was the sender (iperf3 -c 192.168.4.153 -b 40M -t 10s). After I reversed them today, the test results did not change.

It is not the issue with the ECU or chip.
As TCP Is handshake mechanism, iperf application iself not pumping the data while negotiation with client, also you can observe the retries in case of TCP.
If you are sending custom TCP packet without using standard applications like iperf then you can able to observe the rate liming from chip.

Lei Zhang said:

First, why is it that I set pir to 20Mb,cir to 10Mb, but the test result via iperf3 about udp is only about 6Mb. Isn't the difference too big?

Can you check the throughput using the CPSW stats it might be close to 10Mbps + 1/2Mbps as these are bits per second but drop will be in packets, If you are sending small size packets with then you may be observe the throughput close to it.
Also, even when you are setting pir 20Mb the frames after 10Mbps are marked as yellow, as per your setting yellow also dropped so your throughput is limited to 10Mbps (10M bits per second)
Please refer to below from TRM.



Best Regards,
Sudheer

Hi,

I've done other chip speed limits before and haven't had problems like tcp.

Doredla Sudheer Kumar said:

If you are sending custom TCP packet without using standard applications like iperf then you can able to observe the rate liming from chip.

Can you recommend a software to test tcp?

I tested udp using the iperf3 parameter -l, changing the packet size. but the best test result was about 7M. This is still a bit different from the cir 10M. Can you recommend the parameters that should be set for -l (iperf3 --c 192.168.4.153 -b 40M -t 10s -l 1000) ?

The test result of (iperf3 --c 192.168.4.153 -b 40M -t 10s -l 64):

The test result of (iperf3 --c 192.168.4.153 -b 40M -t 10s -l 200):

The test reulst of (iperf3 --c 192.168.4.153 -b 40M -t 10s -l 1000):

Best Regards,
Lei

Hi,

Lei Zhang said:

I've done other chip speed limits before and haven't had problems like tcp.

Have you used similar test environment for testing the same. If so, kindly check whether flow control might be enabled in that case or not.

Lei Zhang said:

Doredla Sudheer Kumar said:

If you are sending custom TCP packet without using standard applications like iperf then you can able to observe the rate liming from chip.

Can you recommend a software to test tcp?

What I suggested is pushing more TCP traffic, which is as good as pushing any traffic as the processing of all types frames will be same at H/W and S/W only application it will be differ based on usage.
As we see UDP case we are almost getting 8Mbps throughput.

Lei Zhang said:

I tested udp using the iperf3 parameter -l, changing the packet size. but the best test result was about 7M. This is still a bit different from the cir 10M. Can you recommend the parameters that should be set for -l (iperf3 --c 192.168.4.153 -b 40M -t 10s -l 1000) ?

As we can see 8Mbps is achieving with UDP, as per internal discussion dropping of packets not at exactly choppy when huge data is receiving so, you may get some difference.
Also, you can check by sending traffic with less bandwidth request like "-b 15M" you may get better result.

Also, can you perform report using the CPSW statistics similar to what I have shared earlier.
Please refer to FAQ [ How to collect /get CPSW Ethernet statistics].


Best Regards,
Sudheer

Hi,

Doredla Sudheer Kumar said:

Also, you can check by sending traffic with less bandwidth request like "-b 15M" you may get better result.

I have tested it and the result is as follow:

The first picture is tda4vm and the second is another ecu. The statistic is as follow:

[MCU2_0]  Port 0 Statistics
[MCU2_0]   2611.987818 s: -----------------------------------------
[MCU2_0]   2611.987869 s:   rxGoodFrames            = 498
[MCU2_0]   2611.987915 s:   rxBcastFrames           = 47
[MCU2_0]   2611.987953 s:   rxMcastFrames           = 272
[MCU2_0]   2611.987990 s:   aleDrop                 = 421
[MCU2_0]   2611.988049 s:   rxOctets                = 42976
[MCU2_0]   2611.988088 s:   txGoodFrames            = 124480
[MCU2_0]   2611.988129 s:   txOctets                = 149971126
[MCU2_0]   2611.988167 s:   octetsFrames64          = 77
[MCU2_0]   2611.988203 s:   octetsFrames65to127     = 471
[MCU2_0]   2611.988239 s:   octetsFrames128to255    = 17
[MCU2_0]   2611.988275 s:   octetsFrames256to511    = 6
[MCU2_0]   2611.988312 s:   octetsFrames1024        = 124407
[MCU2_0]   2611.988352 s:   netOctets               = 150014102
[MCU2_0]   2611.988390 s:   portMaskDrop            = 421
[MCU2_0]   2611.988427 s:   aleVidIngressDrop       = 377
[MCU2_0]   2611.988463 s:   aleUnknownUcast         = 22
[MCU2_0]   2611.988500 s:   aleUnknownUcastBcnt     = 1804
[MCU2_0]   2611.988536 s:   aleUnknownMcast         = 19
[MCU2_0]   2611.988571 s:   aleUnknownMcastBcnt     = 1633
[MCU2_0]   2611.988614 s:   txPri[0]                = 124480
[MCU2_0]   2611.988656 s:   txPriBcnt[0]            = 149971126
[MCU2_0]   2611.988684 s:
[MCU2_0]   2611.988756 s:
[MCU2_0]  External Port 0 Statistics
[MCU2_0]   2611.988780 s: -----------------------------------------
[MCU2_0]   2611.988815 s:
[MCU2_0]   2611.988879 s:
[MCU2_0]  External Port 1 Statistics
[MCU2_0]   2611.988904 s: -----------------------------------------
[MCU2_0]   2611.988931 s:
[MCU2_0]   2611.988989 s:
[MCU2_0]  External Port 2 Statistics
[MCU2_0]   2611.989038 s: -----------------------------------------
[MCU2_0]   2611.989091 s:   rxGoodFrames            = 126254
[MCU2_0]   2611.989133 s:   rxBcastFrames           = 28
[MCU2_0]   2611.989169 s:   rxMcastFrames           = 1438
[MCU2_0]   2611.989205 s:   aleDrop                 = 1774
[MCU2_0]   2611.989243 s:   rxOctets                = 150136733
[MCU2_0]   2611.989279 s:   txGoodFrames            = 77
[MCU2_0]   2611.989313 s:   txMcastFrames           = 10
[MCU2_0]   2611.989351 s:   txOctets                = 7642
[MCU2_0]   2611.989388 s:   octetsFrames64          = 49
[MCU2_0]   2611.989423 s:   octetsFrames65to127     = 1802
[MCU2_0]   2611.989456 s:   octetsFrames128to255    = 67
[MCU2_0]   2611.989489 s:   octetsFrames256to511    = 6
[MCU2_0]   2611.989524 s:   octetsFrames1024        = 124407
[MCU2_0]   2611.989560 s:   netOctets               = 150144375
[MCU2_0]   2611.989597 s:   portMaskDrop            = 1774
[MCU2_0]   2611.989632 s:   aleVidIngressDrop       = 1751
[MCU2_0]   2611.989667 s:   aleUnknownUcast         = 269
[MCU2_0]   2611.989702 s:   aleUnknownUcastBcnt     = 21794
[MCU2_0]   2611.989738 s:   aleUnknownMcast         = 785
[MCU2_0]   2611.989775 s:   aleUnknownMcastBcnt     = 75469
[MCU2_0]   2611.989809 s:   aleUnknownBcast         = 20
[MCU2_0]   2611.989844 s:   aleUnknownBcastBcnt     = 1280
[MCU2_0]   2611.989879 s:   alePolicyMatch          = 124478
[MCU2_0]   2611.989915 s:   txPri[0]                = 77
[MCU2_0]   2611.989951 s:   txPriBcnt[0]            = 7642
[MCU2_0]   2611.989974 s:
[MCU2_0]   2611.990056 s:
[MCU2_0]  External Port 3 Statistics
[MCU2_0]   2611.990085 s: -----------------------------------------
[MCU2_0]   2611.990113 s:
[MCU2_0]   2611.990174 s:
[MCU2_0]  External Port 4 Statistics
[MCU2_0]   2611.990196 s: -----------------------------------------
[MCU2_0]   2611.990222 s:
[MCU2_0]   2611.990281 s:
[MCU2_0]  External Port 5 Statistics
[MCU2_0]   2611.990302 s: -----------------------------------------
[MCU2_0]   2611.990329 s:
[MCU2_0]   2611.990386 s:
[MCU2_0]  External Port 6 Statistics
[MCU2_0]   2611.990406 s: -----------------------------------------
[MCU2_0]   2611.990432 s:
[MCU2_0]   2611.990489 s:
[MCU2_0]  External Port 7 Statistics
[MCU2_0]   2611.990509 s: -----------------------------------------
[MCU2_0]   2611.990534 s:

After the test of iperf3:

[MCU2_0]  Port 0 Statistics
[MCU2_0]   2659.907760 s: -----------------------------------------
[MCU2_0]   2659.907813 s:   rxGoodFrames            = 541
[MCU2_0]   2659.907858 s:   rxBcastFrames           = 47
[MCU2_0]   2659.907895 s:   rxMcastFrames           = 274
[MCU2_0]   2659.907931 s:   aleDrop                 = 446
[MCU2_0]   2659.907967 s:   rxOctets                = 46727
[MCU2_0]   2659.908004 s:   txGoodFrames            = 134001
[MCU2_0]   2659.908064 s:   txMcastFrames           = 1
[MCU2_0]   2659.908103 s:   txOctets                = 159949091
[MCU2_0]   2659.908142 s:   octetsFrames64          = 79
[MCU2_0]   2659.908177 s:   octetsFrames65to127     = 528
[MCU2_0]   2659.908213 s:   octetsFrames128to255    = 19
[MCU2_0]   2659.908247 s:   octetsFrames256to511    = 8
[MCU2_0]   2659.908284 s:   octetsFrames1024        = 133908
[MCU2_0]   2659.908324 s:   netOctets               = 159995818
[MCU2_0]   2659.908361 s:   portMaskDrop            = 446
[MCU2_0]   2659.908397 s:   aleVidIngressDrop       = 402
[MCU2_0]   2659.908433 s:   aleUnknownUcast         = 22
[MCU2_0]   2659.908469 s:   aleUnknownUcastBcnt     = 1804
[MCU2_0]   2659.908503 s:   aleUnknownMcast         = 19
[MCU2_0]   2659.908538 s:   aleUnknownMcastBcnt     = 1633
[MCU2_0]   2659.908580 s:   txPri[0]                = 134001
[MCU2_0]   2659.908621 s:   txPriBcnt[0]            = 159949091
[MCU2_0]   2659.908648 s:
[MCU2_0]   2659.908717 s:
[MCU2_0]  External Port 0 Statistics
[MCU2_0]   2659.908742 s: -----------------------------------------
[MCU2_0]   2659.908778 s:
[MCU2_0]   2659.908841 s:
[MCU2_0]  External Port 1 Statistics
[MCU2_0]   2659.908862 s: -----------------------------------------
[MCU2_0]   2659.908888 s:
[MCU2_0]   2659.908944 s:
[MCU2_0]  External Port 2 Statistics
[MCU2_0]   2659.908967 s: -----------------------------------------
[MCU2_0]   2659.909030 s:   rxGoodFrames            = 145268
[MCU2_0]   2659.909079 s:   rxBcastFrames           = 32
[MCU2_0]   2659.909116 s:   rxMcastFrames           = 1640
[MCU2_0]   2659.909152 s:   aleDrop                 = 2019
[MCU2_0]   2659.909190 s:   rxOctets                = 169847796
[MCU2_0]   2659.909227 s:   txGoodFrames            = 95
[MCU2_0]   2659.909261 s:   txMcastFrames           = 11
[MCU2_0]   2659.909299 s:   txOctets                = 9347
[MCU2_0]   2659.909335 s:   octetsFrames64          = 57
[MCU2_0]   2659.909370 s:   octetsFrames65to127     = 2065
[MCU2_0]   2659.909407 s:   octetsFrames128to255    = 77
[MCU2_0]   2659.909440 s:   octetsFrames256to511    = 8
[MCU2_0]   2659.909474 s:   octetsFrames1024        = 143156
[MCU2_0]   2659.909511 s:   netOctets               = 169857143
[MCU2_0]   2659.909548 s:   portMaskDrop            = 11267
[MCU2_0]   2659.909583 s:   aleVidIngressDrop       = 1996
[MCU2_0]   2659.909620 s:   aleUnknownUcast         = 299
[MCU2_0]   2659.909656 s:   aleUnknownUcastBcnt     = 24134
[MCU2_0]   2659.909691 s:   aleUnknownMcast         = 825
[MCU2_0]   2659.909727 s:   aleUnknownMcastBcnt     = 79197
[MCU2_0]   2659.909762 s:   aleUnknownBcast         = 22
[MCU2_0]   2659.909797 s:   aleUnknownBcastBcnt     = 1408
[MCU2_0]   2659.909832 s:   alePolicyMatch          = 143246
[MCU2_0]   2659.909867 s:   alePolicyMatchRed       = 8752
[MCU2_0]   2659.909902 s:   alePolicyMatchYellow    = 496
[MCU2_0]   2659.909939 s:   txPri[0]                = 95
[MCU2_0]   2659.909976 s:   txPriBcnt[0]            = 9347
[MCU2_0]   2659.910000 s:
[MCU2_0]   2659.910082 s:
[MCU2_0]  External Port 3 Statistics
[MCU2_0]   2659.910111 s: -----------------------------------------
[MCU2_0]   2659.910142 s:
[MCU2_0]   2659.910205 s:
[MCU2_0]  External Port 4 Statistics
[MCU2_0]   2659.910227 s: -----------------------------------------
[MCU2_0]   2659.910255 s:
[MCU2_0]   2659.910314 s:
[MCU2_0]  External Port 5 Statistics
[MCU2_0]   2659.910336 s: -----------------------------------------
[MCU2_0]   2659.910362 s:
[MCU2_0]   2659.910422 s:
[MCU2_0]  External Port 6 Statistics
[MCU2_0]   2659.910443 s: -----------------------------------------
[MCU2_0]   2659.910469 s:
[MCU2_0]   2659.910528 s:
[MCU2_0]  External Port 7 Statistics
[MCU2_0]   2659.910549 s: -----------------------------------------
[MCU2_0]   2659.910575 s:

Best Regards,
Lei

Hi,

I could see ~8Mbps throughput from the CPSW statistics, also some Port Mask drop in your case.
Please refer to below.


Best Regards,
Sudheer

Hi,

The rate limit is 10Mbps but the test result is 8Mbps through the CPSW statistics. I think the gap is is still a bit wide. Is there something wrong with my settings, or is there something wrong with this sdk0703? 

Doredla Sudheer Kumar said:

some Port Mask drop in your case.

I don't know why some Port Mask drop in my case. Even with the addition of these few dropped packets, this rate is still about 8Mbps.

Best Regards,
Lei

Hi,

Sorry for delay.
We are confirming internally with experts about the gap in policer rate limiting and get back to you on this.

Best Regards,
Sudheer

Hi,

why it's 50.5%. If the ingress rate is 10% above the expected rate, is the rate 60% of the limit? I don't really understand the logic.

Doredla Sudheer Kumar said:

. if your ingress rate is 1% above the expected rate we would drop every other packet, resulting in 50.5% of the limit.

Best regards,

Lei

Hi, 

Above is the HW behavior, and it was same as IEEE & IET rate limiting process.

Best Regards, 

Sudheer