test suite reviews and discussions
 help / color / mirror / Atom feed
From: "Liu, Yong" <yong.liu@intel.com>
To: "Xu, Qian Q" <qian.q.xu@intel.com>, "dts@dpdk.org" <dts@dpdk.org>
Subject: Re: [dts] [PATCH 1/2][dts v2] Update vmdq test scripts for supporting	FVL/Niantic.
Date: Wed, 16 Sep 2015 02:45:38 +0000	[thread overview]
Message-ID: <86228AFD5BCD8E4EBFD2B90117B5E81E10EF46BF@SHSMSX103.ccr.corp.intel.com> (raw)
In-Reply-To: <1442371032-25142-1-git-send-email-qian.q.xu@intel.com>

Thanks Qian. Applied.

> -----Original Message-----
> From: dts [mailto:dts-bounces@dpdk.org] On Behalf Of Qian Xu
> Sent: Wednesday, September 16, 2015 10:37 AM
> To: dts@dpdk.org
> Subject: [dts] [PATCH 1/2][dts v2] Update vmdq test scripts for supporting
> FVL/Niantic.
> 
> [v2]
> Minor changes on the code format and value assignment.
> [v1]
> Updated vmdq test suite, and it works for both FVL and Niantic. The major
> changes are as below:
> 1. Support FVL 4x10g, 2x10g, 2x40g.
> 2. Update the traffic to make the MAC address and VLAN id changes together.
> Then the packet loss is 0 at low traffic.
> 3. Set the max pools number for major NICs: FVL 40G, 10G, Niantic, 1G.
> 4. Update the core config.
> 
> Signed-off-by: Qian Xu <qian.q.xu@intel.com>
> 
> diff --git a/tests/TestSuite_vmdq.py b/tests/TestSuite_vmdq.py
> new file mode 100644
> index 0000000..de5a86f
> --- /dev/null
> +++ b/tests/TestSuite_vmdq.py
> @@ -0,0 +1,262 @@
> +# <COPYRIGHT_TAG>
> +
> +"""
> +DPDK Test suite.
> +
> +Tests for vmdq.
> +
> +"""
> +
> +import dts
> +import re
> +from etgen import IxiaPacketGenerator
> +from test_case import TestCase
> +from time import sleep
> +
> +
> +class TestVmdq(TestCase, IxiaPacketGenerator):
> +    dut_ports = []
> +    ip_dot1q_header_size = 22
> +    default_framesize = 64
> +    default_payload = default_framesize - ip_dot1q_header_size
> +    current_frame_size = 0
> +    destmac_port0 = "52:54:00:12:00:00"
> +    destmac_port1 = "52:54:00:12:01:00"
> +    da_repeat = 1
> +    vlan_repeat = 1
> +    queues = 8
> +
> +    def set_up_all(self):
> +        """
> +        Run at the start of each test suite.
> +        """
> +
> +        self.tester.extend_external_packet_generator(TestVmdq, self)
> +
> +        self.dut.send_expect("sed -i
> 's/CONFIG_RTE_MAX_QUEUES_PER_PORT=256/CONFIG_RTE_MAX_QUEUES_PER_PORT=1024/
> ' ./config/common_linuxapp", "# ", 5)
> +
> +        self.dut.build_install_dpdk(self.target)
> +        # Update the max queue per port for Fortville.
> +        self.dut.send_expect("sed -i 's/define MAX_QUEUES 128/define
> MAX_QUEUES 1024/' ./examples/vmdq/main.c", "#", 5)
> +
> +        self.dut_ports = self.dut.get_ports(self.nic)
> +        print self.dut_ports
> +        self.verify(len(self.dut_ports) >= 2, "Insufficient ports")
> +
> +        self.core_configs = []
> +        self.core_configs.append({'cores': '1S/1C/1T', 'mpps': {}})
> +        self.core_configs.append({'cores': '1S/2C/1T', 'mpps': {}})
> +        self.core_configs.append({'cores': '1S/2C/2T', 'mpps': {}})
> +        self.core_configs.append({'cores': '1S/4C/1T', 'mpps': {}})
> +
> +        self.ports_socket = self.dut.get_numa_id(self.dut_ports[0])
> +        out = self.dut.send_expect("make -C examples/vmdq", "#", 10)
> +        self.verify("Error" not in out, "Compilation error")
> +
> +
> +    def validateApproxEqual(self, lines):
> +        """
> +        Check that all the rx queue stats are within a 30% range.
> +        """
> +
> +        minimum = 1000000
> +        maximun = 0
> +
> +        # Need to use Python re package because dts.regexp only handles 1
> group,
> +        # we need 4.
> +        scanner = re.compile(
> +            "^Pool [0-9]+: ([0-9]+) ([0-9]+) ([0-9]+) ([0-9]+)$")
> +        for l in lines:
> +            m = scanner.search(l)
> +            if m is None:
> +                # Line at the end, "Finished handling signal", ignore
> +                pass
> +            else:
> +                for stat in m.groups():
> +                    if stat < minimum:
> +                        minimum = stat
> +                    if stat > maximun:
> +                        maximun = stat
> +        self.verify(maximun - minimum <= minimum *
> +                    0.3, "Too wide variation in queue stats")
> +
> +    def Npools_128queues(self, npools):
> +        """
> +        MAX queues is 128
> +        queues/pools = 128/npools
> +        """
> +
> +        self.current_frame_size = self.default_framesize
> +
> +        self.dut_ports = self.dut.get_ports(self.nic)
> +
> +        core_list = self.dut.get_core_list("1S/4C/1T",
> socket=self.ports_socket)
> +        core_mask = dts.create_mask(core_list)
> +
> +        port_mask = dts.create_mask([self.dut_ports[0],
> self.dut_ports[1]])
> +        # Run the application
> +        out = self.dut.send_expect("./examples/vmdq/build/vmdq_app -n 4 -
> c %s -- -p %s --nb-pools %s&" %
> +                                   (core_mask, port_mask, str(npools)),
> "reading queues", 120)
> +
> +        # Transmit traffic
> +        tx_port = self.tester.get_local_port(self.dut_ports[0])
> +        rx_port = self.tester.get_local_port(self.dut_ports[1])
> +        tx_mac = self.tester.get_mac(tx_port)
> +
> +        self.vlan_repeat = npools
> +        self.da_repeat = npools
> +        tgen_input = []
> +        for p in range(8):
> +            self.tester.scapy_append('dmac="%s"' % self.destmac_port0)
> +            self.tester.scapy_append('smac="%s"' % tx_mac)
> +            self.tester.scapy_append(
> +                'flows = [Ether(src=smac,
> dst=dmac)/Dot1Q(vlan=0,prio=%d)]'%p)
> +            self.tester.scapy_append('wrpcap("test%d.pcap", flows)' %p)
> +            self.tester.scapy_execute()
> +            tgen_input.append((tx_port, rx_port, "test%d.pcap" %p))
> +
> +        loss = self.tester.traffic_generator_loss(tgen_input, 10)
> +        print "loss is %s !" % loss
> +
> +        # Verify the accurate
> +        self.verify(loss < 0.001, "Excessive packet loss")
> +        self.validateApproxEqual(out.split("\r\n"))
> +
> +    def set_up(self):
> +        """
> +        Run before each test case.
> +        """
> +        self.dut.kill_all()
> +
> +    def test_perf_vmdq_64pools_queues(self):
> +        """
> +        This function call " Npools_128queues" with differen number
> +        of pools. Details see below. if not sure, set it as 8 pools.
> +        """
> +        if self.nic in ("niantic", "springfountain"):
> +            self.Npools_128queues(64)
> +        elif self.nic in ("fortville_spirit", "fortville_spirit_single"):
> +            self.Npools_128queues(63)
> +        elif self.nic in ("fortville_eagle"):
> +            self.Npools_128queues(34)
> +        else:
> +            self.Npools_128queues(8)
> +
> +    def test_perf_vmdq_performance(self):
> +        """
> +        Try  different configuration and different packe size
> +        """
> +
> +        self.tester.get_interface(
> +            self.tester.get_local_port(self.dut_ports[0]))
> +
> +        frame_sizes = [64, 128, 256, 512, 1024, 1280, 1518]
> +        for config in self.core_configs:
> +
> +            print dts.BLUE(config["cores"])
> +            self.dut.kill_all()
> +
> +            core_config = config['cores']
> +            core_list =
> self.dut.get_core_list(core_config,socket=self.ports_socket)
> +            core_mask = dts.create_mask(core_list)
> +            portmask = dts.create_mask(self.dut.get_ports())
> +            if self.nic in ("niantic", "springfountain"):
> +                self.queues = 64
> +                self.dut.send_expect(
> +                    "examples/vmdq/build/vmdq_app -n %d -c %s -- -p %s --
> nb-pools 64&" %
> +                    (self.dut.get_memory_channels(), core_mask, portmask),
> "reading queues", 30)
> +            elif self.nic in ("fortville_spirit",
> "fortville_spirit_single"):
> +                self.queues = 63
> +                self.dut.send_expect(
> +                    "examples/vmdq/build/vmdq_app -n %d -c %s -- -p %s --
> nb-pools 63&" %
> +                    (self.dut.get_memory_channels(), core_mask, portmask),
> "reading queues", 30)
> +            elif self.nic in ("fortville_eagle"):
> +                self.queues = 34
> +                self.dut.send_expect(
> +                    "examples/vmdq/build/vmdq_app -n %d -c %s -- -p %s --
> nb-pools 34&" %
> +                    (self.dut.get_memory_channels(), core_mask, portmask),
> "reading queues", 30)
> +            else:
> +                self.queues = 8
> +                self.dut.send_expect(
> +                    "examples/vmdq/build/vmdq_app -n %d -c %s -- -p %s --
> nb-pools 8&" %
> +                    (self.dut.get_memory_channels(), core_mask, portmask),
> "reading queues", 30)
> +
> +            tx_port = self.tester.get_local_port(self.dut_ports[0])
> +            rx_port = self.tester.get_local_port(self.dut_ports[1])
> +
> +            print dts.GREEN("Waiting for application to initialize")
> +            sleep(5)
> +
> +            for frame_size in frame_sizes:
> +
> +                TestVmdq.current_frame_size = frame_size
> +
> +                print dts.BLUE(str(frame_size))
> +
> +                self.tester.scapy_append('dstmac="%s"' %
> self.destmac_port0)
> +                tx_mac = self.tester.get_mac(tx_port)
> +                self.tester.scapy_append('srcmac="%s"' % tx_mac)
> +                self.tester.scapy_append(
> +                        'flows =
> [Ether(src=srcmac,dst=dstmac)/Dot1Q(vlan=0)/("X"*%d)]' %
> +                        (frame_size - TestVmdq.ip_dot1q_header_size))
> +                self.tester.scapy_append('wrpcap("test1.pcap", flows)')
> +                self.tester.scapy_execute()
> +
> +                self.tester.scapy_append('dstmac="%s"' %
> self.destmac_port1)
> +                tx_mac = self.tester.get_mac(rx_port)
> +                self.tester.scapy_append('srcmac="%s"' % tx_mac)
> +                self.tester.scapy_append(
> +                        'flows =
> [Ether(src=srcmac,dst=dstmac)/Dot1Q(vlan=0)/("X"*%d)]' %
> +                        (frame_size - TestVmdq.ip_dot1q_header_size))
> +                self.tester.scapy_append('wrpcap("test2.pcap", flows)')
> +                self.tester.scapy_execute()
> +
> +                self.vlan_repeat = self.queues
> +                self.da_repeat = self.queues
> +
> +                tgen_input = []
> +                tgen_input.append((tx_port, rx_port, "test1.pcap"))
> +                tgen_input.append((rx_port, tx_port, "test2.pcap"))
> +                _, pps =
> self.tester.traffic_generator_throughput(tgen_input)
> +                config['mpps'][frame_size] = pps/1000000.0
> +
> +        for n in range(len(self.core_configs)):
> +            for size in frame_sizes:
> +                self.verify(
> +                    self.core_configs[n]['mpps'][size] is not 0, "No
> traffic detected")
> +
> +        # Print results
> +        dts.results_table_add_header(
> +            ['Frame size'] + [n['cores'] for n in self.core_configs])
> +
> +        for size in frame_sizes:
> +            dts.results_table_add_row(
> +                [size] + [n['mpps'][size] for n in self.core_configs])
> +
> +        dts.results_table_print()
> +
> +    # Override etgen.dot1q function
> +    def dot1q(self, port, prio, id, vlan, type):
> +        """
> +        Change Ixia configuration
> +        """
> +
> +        self.add_tcl_cmd("vlan config -mode vIncrement")
> +        self.add_tcl_cmd("vlan config -step 1")
> +        self.add_tcl_cmd("vlan config -repeat %d" % self.vlan_repeat)
> +        self.add_tcl_cmd("stream config -framesize %d" %
> +                         TestVmdq.current_frame_size)
> +        super(TestVmdq, self).dot1q(port, prio, id, vlan, type)
> +
> +    def ether(self, port, src, dst, type):
> +        """
> +        Configure Ether protocal.
> +        """
> +        self.add_tcl_cmd("protocol config -ethernetType ethernetII")
> +        self.add_tcl_cmd('stream config -sa "%s"' %
> self.macToTclFormat(src))
> +        self.add_tcl_cmd('stream config -da "%s"' %
> self.macToTclFormat(dst))
> +        self.add_tcl_cmd('stream config -daRepeatCounter increment')
> +        self.add_tcl_cmd('stream config -daStep 1')
> +        self.add_tcl_cmd('stream config -numDA %d' % self.da_repeat)
> +
> +
> --
> 2.1.0

      parent reply	other threads:[~2015-09-16  2:45 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-09-16  2:37 Qian Xu
2015-09-16  2:37 ` [dts] [PATCH 2/2][dts v2] Update vmdq test plan to align with scripts Qian Xu
2015-09-16  2:45 ` Liu, Yong [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=86228AFD5BCD8E4EBFD2B90117B5E81E10EF46BF@SHSMSX103.ccr.corp.intel.com \
    --to=yong.liu@intel.com \
    --cc=dts@dpdk.org \
    --cc=qian.q.xu@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).