Re: [EXT] Re: [PATCH net-next 3/5] net: mvpp2: cls: Use RSS contexts to handle RSS tables
From: Russell King - ARM Linux admin
Date: Wed May 20 2020 - 07:11:05 EST
On Tue, May 19, 2020 at 07:05:34PM +0200, Matteo Croce wrote:
> On Tue, 19 May 2020 12:05:20 +0200
> Matteo Croce <mcroce@xxxxxxxxxx> wrote:
>
> Hi,
>
> The patch seems to work. I'm generating traffic with random MAC and IP
> addresses, to have many flows:
>
> # tcpdump -tenni eth2
> 9a:a9:b1:3a:b1:6b > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.4.0 > 192.168.0.4.0: UDP, length 12
> 9e:92:fd:f8:7f:0a > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.4.0 > 192.168.0.4.0: UDP, length 12
> 66:b7:11:8a:c2:1f > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.1.0 > 192.168.0.1.0: UDP, length 12
> 7a:ba:58:bd:9a:62 > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.1.0 > 192.168.0.1.0: UDP, length 12
> 7e:78:a9:97:70:3a > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.2.0 > 192.168.0.2.0: UDP, length 12
> b2:81:91:34:ce:42 > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.2.0 > 192.168.0.2.0: UDP, length 12
> 2a:05:52:d0:d9:3f > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.3.0 > 192.168.0.3.0: UDP, length 12
> ee:ee:47:35:fa:81 > 00:51:82:11:22:02, ethertype IPv4 (0x0800), length 60: 10.0.0.3.0 > 192.168.0.3.0: UDP, length 12
>
> This is the default rate, with rxhash off:
>
> # utraf eth2
> tx: 0 bps 0 pps rx: 397.4 Mbps 827.9 Kpps
> tx: 0 bps 0 pps rx: 396.3 Mbps 825.7 Kpps
> tx: 0 bps 0 pps rx: 396.6 Mbps 826.3 Kpps
> tx: 0 bps 0 pps rx: 396.5 Mbps 826.1 Kpps
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 9 root 20 0 0 0 0 R 99.7 0.0 7:02.58 ksoftirqd/0
> 15 root 20 0 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/1
> 20 root 20 0 0 0 0 S 0.0 0.0 2:01.48 ksoftirqd/2
> 25 root 20 0 0 0 0 S 0.0 0.0 0:32.86 ksoftirqd/3
>
> and this with rx hashing enabled:
>
> # ethtool -K eth2 rxhash on
> # utraf eth2
> tx: 0 bps 0 pps rx: 456.4 Mbps 950.8 Kpps
> tx: 0 bps 0 pps rx: 458.4 Mbps 955.0 Kpps
> tx: 0 bps 0 pps rx: 457.6 Mbps 953.3 Kpps
> tx: 0 bps 0 pps rx: 462.2 Mbps 962.9 Kpps
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 20 root 20 0 0 0 0 R 0.7 0.0 2:02.34 ksoftirqd/2
> 25 root 20 0 0 0 0 S 0.3 0.0 0:33.25 ksoftirqd/3
> 9 root 20 0 0 0 0 S 0.0 0.0 7:52.57 ksoftirqd/0
> 15 root 20 0 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/1
>
>
> The throughput doesn't increase so much, maybe we hit an HW limit of
> the gigabit port. The interesting thing is how the global CPU usage
> drops from 25% to 1%.
> I can't explain this, it could be due to the reduced contention?
Hi Matteo,
Can I take that as a Tested-by ?
Thanks.
--
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC for 0.8m (est. 1762m) line in suburbia: sync at 13.1Mbps down 424kbps up