2.6Mpps with RecordTimeStamp() environment. Normal?

[kthkaya]

Following is the single pipeline packet generator

Idle -> EnsureDPDKBuffer -> Discard;
InfiniteSource(LENGTH 10, LIMIT 10000000, STOP true)
        -> NumberPacket(OFFSET 0)
        -> UDPIP6Encap(2001:2001:2001:2001::1, 1234, 2001:2001:2001:2001::2, 1234)
        -> EtherEncap(0x86DD, 00:04:23:D0:93:63, 00:17:cb:0d:f8:db)
        -> record :: RecordTimestamp()
        -> CheckNumberPacket(OFFSET 62, COUNT 10000000)
        -> diff :: TimestampDiff(OFFSET 62, RECORDER record)
        -> counter :: AverageCounter()
        -> Discard;

Yielding these results

diff.average:
0.0118919
counter.count:
10000000
counter.rate:
2661698.16343

Running on

Model name:          Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz
Stepping:            4
CPU MHz:             2800.103
CPU max MHz:         3600.0000
CPU min MHz:         1200.0000
BogoMIPS:            5600.20

No matter how many cores I pass, I get approximately 2.6Mpps as shown above. If I comment out the etherencap and udpencap, it is approx 3Mpps.

Is this expected?

[tbarbette]

Without the replay it looks more or less normal. InfiniteSource needs probably some more HEADROOM to avoid packet copy in the encap elements. The packets are actually created from scratch and maybe even copied once in this setup. Replay will keep them in memory. Before REPLAY it is also a good thing to have EnsureDPDKBuffer to avoid a copy when sending the packet (using QUICK_CLONE 1 is even best).

[kthkaya]

Thanks Tom! After reading https://github.com/kohler/click/wiki/FAQs-about-click I was hoping to catch something around 7Mpps but obviously, not every element is suitable.
I still have to figure out the latency caused by replay from the previous issue. I will keep you posted.

[tbarbette]

This case is very very different. I achieve at least 15MPPS with a single core of a middle range Xeon with a router config. But creating a packet, computing the chzcksumS after IP/UDP encap and copying the content to a DPDK buffer is very heavy. You may gain something by compiling with --enable-dpdk-pool for that last point. If you don't set headroom to InfiniteSource, I also think you have one more packet copy...

[tbarbette]

This case is very very different. I achieve at least 15MPPS with a single core of a middle range Xeon with a router config. But creating a packet, computing the checksumS after IP/UDP encap and copying the content to a DPDK buffer is very heavy. You may gain something by compiling with --enable-dpdk-pool for that last point. If you don't set headroom to InfiniteSource, I also think you have one more packet copy...