Skip to contentSkip to conversation

net/dpdkoin: add new driver

PATCH In review Matteo Croce 5 participants 5 messages Apr 1, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/dpdkoin: add new driver

DPDK polls the packet in a busy loop. This means that CPU constantly spins looking for packets, regardless of the network traffic.
DPDK does this to reduce latency and avoid using interrupts, at expense of
efficiency: this might consume more processing power and generate more heat than needed, potentially increasing the TCO of a DPDK appliance.

Here comes DPDKoin. DPDKoin is a DPDK PMD, which instead of moving packets, mines cryptocurrencies.
DPDKoin just does a few calculations to every poll cycle, this means that when the network is loaded, DPDKoin consumes just a few cycles, but when there is no traffic, DPDKoin mines currencies in a busy loop. The coins are mined with no extra consumption, as the CPU would spin anyway.

This is a system running with two 10G cards running with DPDK:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8223 root 10 -10 260.6g 597516 19100 S 199.8 0.5 0:17.89 testpmd

And this is the same system with a DPDKoin port added:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
9438 root 10 -10 260.6g 598892 19800 S 199.8 0.5 0:19.43 testpmd

As you can see, we don't observe any change in the CPU usage, and besides, the earned coins lowers the appliance TCO.

Just for reference, a test run log follows:

$ sudo testpmd --no-huge -m 1024 --no-pci --vdev eth_dpdkoin0 --vdev eth_dpdkoin1
EAL: Detected 8 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: Selected IOVA mode 'VA'
EAL: Probing VFIO support...
testpmd: create a new mbuf pool <mbuf_pool_socket_0>: n=203456, size=2176, socket=0
testpmd: preferred mempool ops selected: ring_mp_mc
Configuring Port 0 (socket 0)
Port 0: 4A:50:3C:33:55:21
Configuring Port 1 (socket 0)
Port 1: BA:BB:0A:BB:0C:CA
Checking link statuses...
Done
No commandline core given, start packet forwarding
io packet forwarding - ports=2 - cores=1 - streams=2 - NUMA support enabled, MP allocation mode: native
Logical Core 1 (socket 0) forwards packets on 2 streams:
  RX P=0/Q=0 (socket 0) -> TX P=1/Q=0 (socket 0) peer=02:00:00:00:00:01
  RX P=1/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0) peer=02:00:00:00:00:00

io packet forwarding packets/burst=32
nb forwarding cores=1 - nb forwarding ports=2 port 0: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
port 1: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
libbitcoin 0.1: mining on CPU2
libbitcoin 0.1: mining on CPU4

---
 config/common_base                            |   5 +
 drivers/net/Makefile                          |   1 +
 drivers/net/dpdkoin/Makefile                  |  23 +
 drivers/net/dpdkoin/meson.build               |   3 +
 drivers/net/dpdkoin/rte_eth_dpdkoin.c         | 696 ++++++++++++++++++
 .../net/dpdkoin/rte_pmd_dpdkoin_version.map   |   3 +
 mk/rte.app.mk                                 |   1 +
 7 files changed, 732 insertions(+)
 create mode 100644 drivers/net/dpdkoin/Makefile
 create mode 100644 drivers/net/dpdkoin/meson.build
 create mode 100644 drivers/net/dpdkoin/rte_eth_dpdkoin.c
 create mode 100644 drivers/net/dpdkoin/rte_pmd_dpdkoin_version.map
config/common_base +5 -0
index c31175f9d..fafa23006 100644
@@ -1055,6 +1055,11 @@ CONFIG_RTE_LIBRTE_VHOST_DEBUG=n
 #
 CONFIG_RTE_LIBRTE_PMD_VHOST=n
 
+#
+# Compile DPDKOIN
+#
+CONFIG_RTE_LIBRTE_PMD_DPDKOIN=y
+
 #
 # Compile IFC driver
 # To compile, CONFIG_RTE_LIBRTE_VHOST and CONFIG_RTE_EAL_VFIO
drivers/net/Makefile +1 -0
index 4a7f155fc..a7c5d5dbc 100644
@@ -23,6 +23,7 @@ endif
 ifeq ($(CONFIG_RTE_EAL_VFIO)$(CONFIG_RTE_LIBRTE_FSLMC_BUS),yy)
 DIRS-$(CONFIG_RTE_LIBRTE_DPAA2_PMD) += dpaa2
 endif
+DIRS-$(CONFIG_RTE_LIBRTE_PMD_DPDKOIN) += dpdkoin
 DIRS-$(CONFIG_RTE_LIBRTE_E1000_PMD) += e1000
 DIRS-$(CONFIG_RTE_LIBRTE_ENA_PMD) += ena
 DIRS-$(CONFIG_RTE_LIBRTE_ENETC_PMD) += enetc
drivers/net/dpdkoin/Makefile new file +23 -0
new file mode 100644
index 000000000..a8c66f3b2
@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: BSD-3-Clause
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+#
+# library name
+#
+LIB = librte_pmd_dpdkoin.a
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
+LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs
+LDLIBS += -lrte_bus_vdev
+
+EXPORT_MAP := rte_pmd_dpdkoin_version.map
+
+#
+# all source are stored in SRCS-y
+#
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_DPDKOIN) += rte_eth_dpdkoin.c
+
+include $(RTE_SDK)/mk/rte.lib.mk
drivers/net/dpdkoin/meson.build new file +3 -0
new file mode 100644
index 000000000..d64f980aa
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: BSD-3-Clause
+
+sources = files('rte_eth_dpdkoin.c')
drivers/net/dpdkoin/rte_eth_dpdkoin.c new file +696 -0

696 lines not on this page · raw message

drivers/net/dpdkoin/rte_pmd_dpdkoin_version.map new file +3 -0
new file mode 100644
index 000000000..f9f17e4f6
@@ -0,0 +1,3 @@
+DPDK_20.0 {
+	local: *;
+};
mk/rte.app.mk +1 -0
index d295ca0a5..eff58903e 100644
@@ -216,6 +216,7 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_MVPP2_PMD)      += -lrte_pmd_mvpp2
 _LDLIBS-$(CONFIG_RTE_LIBRTE_MVNETA_PMD)     += -lrte_pmd_mvneta
 _LDLIBS-$(CONFIG_RTE_LIBRTE_NFP_PMD)        += -lrte_pmd_nfp
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_NULL)       += -lrte_pmd_null
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_DPDKOIN)    += -lrte_pmd_dpdkoin -lbitcoin
 _LDLIBS-$(CONFIG_RTE_LIBRTE_OCTEONTX2_PMD)  += -lrte_pmd_octeontx2 -lm
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_PCAP)       += -lrte_pmd_pcap -lpcap
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PFE_PMD)        += -lrte_pmd_pfe
Signature · 1 line
2.25.1
#1 lore raw
Quoted 6 lines · Matteo Croce

DPDK polls the packet in a busy loop. This means that CPU constantly spins looking for packets, regardless of the network traffic.
DPDK does this to reduce latency and avoid using interrupts, at expense of
efficiency: this might consume more processing power and generate more heat than needed, potentially increasing the TCO of a DPDK appliance.

Quoted 7 lines

Here comes DPDKoin. DPDKoin is a DPDK PMD, which instead of moving packets, mines cryptocurrencies.
DPDKoin just does a few calculations to every poll cycle, this means that when the network is loaded, DPDKoin consumes just a few cycles, but when there is no traffic, DPDKoin mines currencies in a busy loop. The coins are mined with no extra consumption, as the CPU would spin anyway.

:)

<...>

drivers/net/dpdkoin/rte_eth_dpdkoin.c line 539

Quoted 8 lines
drivers/net/dpdkoin/rte_eth_dpdkoin.c new file +8 -0
+	/* finally assign rx and tx ops */
+	if (packet_copy) {
+		eth_dev->rx_pkt_burst = eth_dpdkoin_copy_rx;
+		eth_dev->tx_pkt_burst = eth_dpdkoin_copy_tx;
+	} else {
+		eth_dev->rx_pkt_burst = eth_dpdkoin_rx;
+		eth_dev->tx_pkt_burst = eth_dpdkoin_tx;
+	}

This is the biggest give away, but still good job :)

#2 lore raw
In reply to Matteo Croce
Quoted 73 lines · Matteo Croce

DPDK polls the packet in a busy loop. This means that CPU constantly spins looking for packets, regardless of the network traffic.
DPDK does this to reduce latency and avoid using interrupts, at expense of
efficiency: this might consume more processing power and generate more heat than needed, potentially increasing the TCO of a DPDK appliance.

Here comes DPDKoin. DPDKoin is a DPDK PMD, which instead of moving packets, mines cryptocurrencies.
DPDKoin just does a few calculations to every poll cycle, this means that when the network is loaded, DPDKoin consumes just a few cycles, but when there is no traffic, DPDKoin mines currencies in a busy loop. The coins are mined with no extra consumption, as the CPU would spin anyway.

This is a system running with two 10G cards running with DPDK:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8223 root 10 -10 260.6g 597516 19100 S 199.8 0.5 0:17.89 testpmd

And this is the same system with a DPDKoin port added:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
9438 root 10 -10 260.6g 598892 19800 S 199.8 0.5 0:19.43 testpmd

As you can see, we don't observe any change in the CPU usage, and besides, the earned coins lowers the appliance TCO.

Just for reference, a test run log follows:

$ sudo testpmd --no-huge -m 1024 --no-pci --vdev eth_dpdkoin0 --vdev eth_dpdkoin1
EAL: Detected 8 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: Selected IOVA mode 'VA'
EAL: Probing VFIO support...
testpmd: create a new mbuf pool <mbuf_pool_socket_0>: n 3456, size!76, socket=0
testpmd: preferred mempool ops selected: ring_mp_mc
Configuring Port 0 (socket 0)
Port 0: 4A:50:3C:33:55:21
Configuring Port 1 (socket 0)
Port 1: BA:BB:0A:BB:0C:CA
Checking link statuses...
Done
No commandline core given, start packet forwarding
io packet forwarding - ports=2 - cores=1 - streams=2 - NUMA support enabled, MP allocation mode: native
Logical Core 1 (socket 0) forwards packets on 2 streams:
  RX P=0/Q=0 (socket 0) -> TX P=1/Q=0 (socket 0) peer:00:00:00:00:01
  RX P=1/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0) peer:00:00:00:00:00

io packet forwarding packets/burst2
nb forwarding cores=1 - nb forwarding ports=2 port 0: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
port 1: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
libbitcoin 0.1: mining on CPU2
libbitcoin 0.1: mining on CPU4

Quoted 14 lines
---
 config/common_base                            |   5 +
 drivers/net/Makefile                          |   1 +
 drivers/net/dpdkoin/Makefile                  |  23 +
 drivers/net/dpdkoin/meson.build               |   3 +
 drivers/net/dpdkoin/rte_eth_dpdkoin.c         | 696 ++++++++++++++++++
 .../net/dpdkoin/rte_pmd_dpdkoin_version.map   |   3 +
 mk/rte.app.mk                                 |   1 +
 7 files changed, 732 insertions(+)
 create mode 100644 drivers/net/dpdkoin/Makefile
 create mode 100644 drivers/net/dpdkoin/meson.build
 create mode 100644 drivers/net/dpdkoin/rte_eth_dpdkoin.c
 create mode 100644 drivers/net/dpdkoin/rte_pmd_dpdkoin_version.map

Is it possible to have a unit test for this PMD? We could have the robot run some quick tests for us... ;-)

#3 lore raw
In reply to Matteo Croce
Quoted 73 lines · Matteo Croce

DPDK polls the packet in a busy loop. This means that CPU constantly spins looking for packets, regardless of the network traffic.
DPDK does this to reduce latency and avoid using interrupts, at expense of
efficiency: this might consume more processing power and generate more heat than needed, potentially increasing the TCO of a DPDK appliance.

Here comes DPDKoin. DPDKoin is a DPDK PMD, which instead of moving packets, mines cryptocurrencies.
DPDKoin just does a few calculations to every poll cycle, this means that when the network is loaded, DPDKoin consumes just a few cycles, but when there is no traffic, DPDKoin mines currencies in a busy loop. The coins are mined with no extra consumption, as the CPU would spin anyway.

This is a system running with two 10G cards running with DPDK:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8223 root 10 -10 260.6g 597516 19100 S 199.8 0.5 0:17.89 testpmd

And this is the same system with a DPDKoin port added:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
9438 root 10 -10 260.6g 598892 19800 S 199.8 0.5 0:19.43 testpmd

As you can see, we don't observe any change in the CPU usage, and besides, the earned coins lowers the appliance TCO.

Just for reference, a test run log follows:

$ sudo testpmd --no-huge -m 1024 --no-pci --vdev eth_dpdkoin0 --vdev eth_dpdkoin1
EAL: Detected 8 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: Selected IOVA mode 'VA'
EAL: Probing VFIO support...
testpmd: create a new mbuf pool <mbuf_pool_socket_0>: n=203456, size=2176, socket=0
testpmd: preferred mempool ops selected: ring_mp_mc
Configuring Port 0 (socket 0)
Port 0: 4A:50:3C:33:55:21
Configuring Port 1 (socket 0)
Port 1: BA:BB:0A:BB:0C:CA
Checking link statuses...
Done
No commandline core given, start packet forwarding
io packet forwarding - ports=2 - cores=1 - streams=2 - NUMA support enabled, MP allocation mode: native
Logical Core 1 (socket 0) forwards packets on 2 streams:
  RX P=0/Q=0 (socket 0) -> TX P=1/Q=0 (socket 0) peer=02:00:00:00:00:01
  RX P=1/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0) peer=02:00:00:00:00:00

io packet forwarding packets/burst=32
nb forwarding cores=1 - nb forwarding ports=2 port 0: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
port 1: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
libbitcoin 0.1: mining on CPU2
libbitcoin 0.1: mining on CPU4

Quoted 13 lines
---
 config/common_base                            |   5 +
 drivers/net/Makefile                          |   1 +
 drivers/net/dpdkoin/Makefile                  |  23 +
 drivers/net/dpdkoin/meson.build               |   3 +
 drivers/net/dpdkoin/rte_eth_dpdkoin.c         | 696 ++++++++++++++++++
 .../net/dpdkoin/rte_pmd_dpdkoin_version.map   |   3 +
 mk/rte.app.mk                                 |   1 +
 7 files changed, 732 insertions(+)
 create mode 100644 drivers/net/dpdkoin/Makefile
 create mode 100644 drivers/net/dpdkoin/meson.build
 create mode 100644 drivers/net/dpdkoin/rte_eth_dpdkoin.c
 create mode 100644 drivers/net/dpdkoin/rte_pmd_dpdkoin_version.map

I see you are requesting this for backport to stable branches. While this is not fixing a specific commit per se, it is clearly fixing a gap whereby it would be unfair for stable users to miss out on any potential bonanza.

Unfortunately there is a blocker due to validation. Normally, we might be concerned by lack of it, but in this case the concern is that it might encourage too much validation and many lengthy stress tests that could impact release dates.

#4 lore raw
Quoted 93 lines · On Wed, 1 Apr 2020 21:51:53 +0100 Kevin Traynor <[email protected]> wrote:
Quoted 73 lines · Matteo Croce

DPDK polls the packet in a busy loop. This means that CPU constantly spins looking for packets, regardless of the network traffic.
DPDK does this to reduce latency and avoid using interrupts, at expense of
efficiency: this might consume more processing power and generate more heat than needed, potentially increasing the TCO of a DPDK appliance.

Here comes DPDKoin. DPDKoin is a DPDK PMD, which instead of moving packets, mines cryptocurrencies.
DPDKoin just does a few calculations to every poll cycle, this means that when the network is loaded, DPDKoin consumes just a few cycles, but when there is no traffic, DPDKoin mines currencies in a busy loop. The coins are mined with no extra consumption, as the CPU would spin anyway.

This is a system running with two 10G cards running with DPDK:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8223 root 10 -10 260.6g 597516 19100 S 199.8 0.5 0:17.89 testpmd

And this is the same system with a DPDKoin port added:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
9438 root 10 -10 260.6g 598892 19800 S 199.8 0.5 0:19.43 testpmd

As you can see, we don't observe any change in the CPU usage, and besides, the earned coins lowers the appliance TCO.

Just for reference, a test run log follows:

$ sudo testpmd --no-huge -m 1024 --no-pci --vdev eth_dpdkoin0 --vdev eth_dpdkoin1
EAL: Detected 8 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: Selected IOVA mode 'VA'
EAL: Probing VFIO support...
testpmd: create a new mbuf pool <mbuf_pool_socket_0>: n=203456, size=2176, socket=0
testpmd: preferred mempool ops selected: ring_mp_mc
Configuring Port 0 (socket 0)
Port 0: 4A:50:3C:33:55:21
Configuring Port 1 (socket 0)
Port 1: BA:BB:0A:BB:0C:CA
Checking link statuses...
Done
No commandline core given, start packet forwarding
io packet forwarding - ports=2 - cores=1 - streams=2 - NUMA support enabled, MP allocation mode: native
Logical Core 1 (socket 0) forwards packets on 2 streams:
  RX P=0/Q=0 (socket 0) -> TX P=1/Q=0 (socket 0) peer=02:00:00:00:00:01
  RX P=1/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0) peer=02:00:00:00:00:00

io packet forwarding packets/burst=32
nb forwarding cores=1 - nb forwarding ports=2 port 0: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
port 1: RX queue number: 1 Tx queue number: 1
Rx offloads=0x0 Tx offloads=0x0
RX queue: 0
RX desc=0 - RX free threshold=0
RX threshold registers: pthresh=0 hthresh=0 wthresh=0 RX Offloads=0x0
TX queue: 0
TX desc=0 - TX free threshold=0
TX threshold registers: pthresh=0 hthresh=0 wthresh=0 TX offloads=0x0 - TX RS bit threshold=0
libbitcoin 0.1: mining on CPU2
libbitcoin 0.1: mining on CPU4

Quoted 13 lines
---
 config/common_base                            |   5 +
 drivers/net/Makefile                          |   1 +
 drivers/net/dpdkoin/Makefile                  |  23 +
 drivers/net/dpdkoin/meson.build               |   3 +
 drivers/net/dpdkoin/rte_eth_dpdkoin.c         | 696 ++++++++++++++++++
 .../net/dpdkoin/rte_pmd_dpdkoin_version.map   |   3 +
 mk/rte.app.mk                                 |   1 +
 7 files changed, 732 insertions(+)
 create mode 100644 drivers/net/dpdkoin/Makefile
 create mode 100644 drivers/net/dpdkoin/meson.build
 create mode 100644 drivers/net/dpdkoin/rte_eth_dpdkoin.c
 create mode 100644 drivers/net/dpdkoin/rte_pmd_dpdkoin_version.map  

I see you are requesting this for backport to stable branches. While this is not fixing a specific commit per se, it is clearly fixing a gap whereby it would be unfair for stable users to miss out on any potential bonanza.

Quoted 5 lines

Unfortunately there is a blocker due to validation. Normally, we might be concerned by lack of it, but in this case the concern is that it might encourage too much validation and many lengthy stress tests that could impact release dates.

Could we fund the CI lab with this?