From: "Pei, Yulong" <yulong.pei@intel.com>
To: "Liu, Yong" <yong.liu@intel.com>, "dts@dpdk.org" <dts@dpdk.org>
Subject: Re: [dts] [PATCH 2/2] tests l3fwd: fix issue for performance test
Date: Thu, 2 Mar 2017 02:49:03 +0000 [thread overview]
Message-ID: <188971FCDA171749BED5DA74ABF3E6F03B690748@shsmsx102.ccr.corp.intel.com> (raw)
In-Reply-To: <86228AFD5BCD8E4EBFD2B90117B5E81E62D6CC33@SHSMSX103.ccr.corp.intel.com>
Hi,
-----Original Message-----
From: Liu, Yong
Sent: Thursday, March 2, 2017 9:54 AM
To: Pei, Yulong <yulong.pei@intel.com>; dts@dpdk.org
Subject: RE: [PATCH 2/2] tests l3fwd: fix issue for performance test
Yulong,
Some questions below.
Thank,
Marvin
> -----Original Message-----
> From: Pei, Yulong
> Sent: Thursday, March 02, 2017 9:36 AM
> To: dts@dpdk.org; Liu, Yong <yong.liu@intel.com>
> Cc: Pei, Yulong <yulong.pei@intel.com>
> Subject: [PATCH 2/2] tests l3fwd: fix issue for performance test
>
> 1. do changes due to dpdk example l3fwd src change.
> 2. remove useless code.
> 3. change calculate method for run_rfc2544 function.
> 4. use tcp packet stream for traffic
>
> Signed-off-by: Yulong Pei <yulong.pei@intel.com>
> ---
> framework/tester.py | 24 ++---
> tests/TestSuite_l3fwd.py | 271
> ++++--------------------------------------
> -----
> 2 files changed, 27 insertions(+), 268 deletions(-)
>
> diff --git a/framework/tester.py b/framework/tester.py index
> 50f687d..d793e91 100644
> --- a/framework/tester.py
> +++ b/framework/tester.py
> @@ -480,30 +480,20 @@ class Tester(Crb):
>
> def run_rfc2544(self, portlist, delay=120, permit_loss_rate=0):
> """
> - zero_rate: dpdk will not lost packet in this line rate.
> - loss_rate: dpdk will loss packet in this line rate.
> test_rate: the line rate we are going to test.
> """
> - zero_rate = 0.0
> - loss_rate = 100.0
> - test_rate = 100.0
> + test_rate = float(100)
>
> - while (loss_rate - zero_rate) > 0.002:
> - self.logger.info("test rate: %f " % test_rate)
> - if test_rate == 100:
> - lost, tx_num, rx_num =
> self.traffic_generator_loss(portlist, test_rate, delay)
> - else:
> - lost, _, _ = self.traffic_generator_loss(portlist,
> test_rate, delay)
> - if lost > permit_loss_rate:
> - loss_rate = test_rate
> - test_rate = (test_rate + zero_rate)/2
> - else:
> - zero_rate = test_rate
> - test_rate = (test_rate + loss_rate)/2
> + self.logger.info("test rate: %f " % test_rate)
> + loss_rate, tx_num, rx_num =
> + self.traffic_generator_loss(portlist,
> test_rate, delay)
> + while loss_rate > permit_loss_rate:
> + test_rate = float(1 - loss_rate) * test_rate
> + loss_rate, tx_num, rx_num =
> self.traffic_generator_loss(portlist, test_rate, delay)
Will this algorithm got same result like bisection algorithm?
Based on the past performance test for Niantic , fvl 10G & 40G, it is more accurate than before.
>
> self.logger.info("zero loss rate is %s" % test_rate)
> return test_rate, tx_num, rx_num
>
> +
> def traffic_generator_loss(self, portList, ratePercent, delay=60):
> """
> Run loss performance test on specified ports.
> diff --git a/tests/TestSuite_l3fwd.py b/tests/TestSuite_l3fwd.py index
> 43af04d..86fc29c 100644
> --- a/tests/TestSuite_l3fwd.py
> +++ b/tests/TestSuite_l3fwd.py
> @@ -93,7 +93,8 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> "{IPv4(13,101,0,0), 24, P3}",
> ]
>
> - frame_sizes = [64, 72, 128, 256, 512, 1024, 2048] # 65, 128
> + frame_sizes = [64, 72, 128, 256, 512, 1024, 1518, 2048] # 65,
> + 128
> +
> methods = ['lpm']#, 'exact']
>
> #
> @@ -114,8 +115,6 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> return '%s' % valports[int(portid)]
>
> #
> - #
> - #
> # Test cases.
> #
>
> @@ -151,13 +150,14 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
>
> pat = re.compile("P([0123])")
> # Update config file and rebuild to get best perf on FVL
> -# self.dut.send_expect("sed -i -e
> 's/CONFIG_RTE_PCI_CONFIG=n/CONFIG_RTE_PCI_CONFIG=y/'
> ./config/common_base", "#", 20)
> -# self.dut.send_expect("sed -i -e
> 's/CONFIG_RTE_PCI_EXTENDED_TAG=.*$/CONFIG_RTE_PCI_EXTENDED_TAG=\"on\"/
> ' ./ config/common_base", "#", 20)
> -# self.dut.build_install_dpdk(self.target)
> -
> + self.dut.send_expect("sed -i -e
> 's/CONFIG_RTE_PCI_CONFIG=n/CONFIG_RTE_PCI_CONFIG=y/'
> ./config/common_base", "#", 20)
As I remember these two options won't affect performance anymore.
Past performance test always with these two options , do you agree just keep it temporary ,
And remove it in the future after do more test for it.
> + self.dut.send_expect("sed -i -e
> 's/CONFIG_RTE_PCI_EXTENDED_TAG=.*$/CONFIG_RTE_PCI_EXTENDED_TAG=\"on\"/
> ' ./ config/common_base", "#", 20)
> + self.dut.send_expect("sed -i -e 's/define
> RTE_TEST_RX_DESC_DEFAULT.*$/define RTE_TEST_RX_DESC_DEFAULT 2048/'
> ./examples/l3fwd/main.c", "#", 20)
> + self.dut.send_expect("sed -i -e 's/define
> RTE_TEST_TX_DESC_DEFAULT.*$/define RTE_TEST_TX_DESC_DEFAULT 2048/'
> ./examples/l3fwd/main.c", "#", 20)
> + self.dut.build_install_dpdk(self.target)
>
> # Prepare long prefix match table, replace P(x) port pattern
> - lpmStr = "static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[]
> = {\\\n"
> + lpmStr = "static struct ipv4_l3fwd_lpm_route
> ipv4_l3fwd_lpm_route_array[] = {\\\n"
> for idx in range(len(TestL3fwd.lpm_table)):
> TestL3fwd.lpm_table[idx] = pat.sub(self.portRepl,
> TestL3fwd.lpm_table[idx])
> lpmStr = lpmStr + ' ' * 4 + TestL3fwd.lpm_table[idx] +
> ",\\\n"
> @@ -165,7 +165,7 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> self.logger.debug(lpmStr)
>
> # Prepare host route table, replace P(x) port pattern
> - exactStr = "static struct ipv4_l3fwd_route
> ipv4_l3fwd_route_array[] = {\\\n"
> + exactStr = "static struct ipv4_l3fwd_em_route
> ipv4_l3fwd_em_route_array[] = {\\\n"
> for idx in range(len(TestL3fwd.host_table)):
> TestL3fwd.host_table[idx] = pat.sub(self.portRepl,
> TestL3fwd.host_table[idx])
> exactStr = exactStr + ' ' * 4 + TestL3fwd.host_table[idx]
> + ",\\\n"
> @@ -173,7 +173,7 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> self.logger.debug(exactStr)
>
> # Compile l3fwd with LPM lookup.
> - self.dut.send_expect(r"sed -i
> '/ipv4_l3fwd_route_array\[\].*{/,/^\}\;/c\\%s' examples/l3fwd/main.c"
> % lpmStr, "# ")
> + self.dut.send_expect(r"sed -i
> '/ipv4_l3fwd_lpm_route_array\[\].*{/,/^\}\;/c\\%s'
> examples/l3fwd/l3fwd_lpm.c" % lpmStr, "# ")
> out = self.dut.build_dpdk_apps("./examples/l3fwd",
> "USER_FLAGS=-
> DAPP_LOOKUP_METHOD=1")
> self.verify("Error" not in out, "compilation error 1")
> self.verify("No such file" not in out, "compilation error 2")
> @@ -183,7 +183,7 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> out = self.dut.send_expect("make clean -C examples/l3fwd", "#
> ")
>
> # Compile l3fwd with hash/exact lookup.
> - self.dut.send_expect(r"sed -i -e
> '/ipv4_l3fwd_route_array\[\].*{/,/^\}\;/c\\%s' examples/l3fwd/main.c"
> % exactStr, "# ")
> + self.dut.send_expect(r"sed -i -e
> '/ipv4_l3fwd_em_route_array\[\].*{/,/^\}\;/c\\%s'
> examples/l3fwd/l3fwd_em.c" % exactStr, "# ")
> out = self.dut.build_dpdk_apps("./examples/l3fwd",
> "USER_FLAGS=-
> DAPP_LOOKUP_METHOD=0")
>
> self.verify("Error" not in out, "compilation error 1") @@
> -226,247 +226,12 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
>
> return '%s,%s,%s' % (str(valports[int(pid)]), qid, lcid)
>
> - def get_throughput(self, frame_size, rx_queues_per_port, cores_config,
> command_line):
> - """
> - Get the throughput for a test case from test_cases_4_ports.
> - """
> -
> - output_pattern = re.compile("P([0123]),([0123]),(C\{\d.\d.\d\})")
> - pat2 = re.compile("C\{\d")
> - repl1 = "C{" + str(self.port_socket)
> -
> - bps = dict()
> - pps = dict()
> - pct = dict()
> -
> - global corelist
> - corelist = []
> -
> -
> - while output_pattern.search(command_line):
> - # If one socket case, we update the socket to ensure the core
> used by l3fwd is on the same socket of the NIC.
> - if cores_config.find('1S')>=0:
> - command_line = pat2.sub(repl1,command_line)
> - command_line = output_pattern.sub(self.repl, command_line)
> -
> -
> - self.logger.debug("%s\n" % str(corelist))
> - core_mask = utils.create_mask(set(corelist))
> -
> - # First, measure by two different methods
> - for method in TestL3fwd.methods:
> - # start l3fwd
> - method_command_line = command_line % (TestL3fwd.path +
> "l3fwd_" + method,
> - core_mask,
> -
> self.dut.get_memory_channels(),
> -
> utils.create_mask(valports[:4]))
> -
> - if frame_size > 1518:
> - method_command_line = method_command_line + " --enable-
> jumbo --max-pkt-len %d" % frame_size
> - self.rst_report(method_command_line + "\n", frame=True,
> annex=True)
> -
> - out = self.dut.send_expect(method_command_line, "L3FWD:", 120)
> -
> - # measure test
> - tgen_input = []
> - for rxPort in range(4):
> - if rxPort % 2 == 0:
> - tx_interface =
> self.tester.get_local_port(valports[rxPort + 1])
> - else:
> - tx_interface =
> self.tester.get_local_port(valports[rxPort - 1])
> -
> - rx_interface =
> self.tester.get_local_port(valports[rxPort])
> - # Make sure the traffic send to the correct MAC address
> - if rxPort % 2 == 0:
> - tgen_input.append((tx_interface, rx_interface,
> "dst%d.pcap" % valports[rxPort+1]))
> - else:
> - tgen_input.append((tx_interface, rx_interface,
> "dst%d.pcap" % valports[rxPort-1]))
> -
> - # FIX ME
> - bps[method], pps[method] =
> self.tester.traffic_generator_throughput(tgen_input)
> - self.verify(pps[method] > 0, "No traffic detected")
> - pps[method] /= 1000000.0
> - pct[method] = pps[method] * 100 /
> float(self.wirespeed(self.nic,
> -
> frame_size,
> - 4))
> -
> - # stop l3fwd
> - self.dut.send_expect("^C", "#")
> -
> - data_row = [frame_size, method, cores_config, rx_queues_per_port]
> - for method in TestL3fwd.methods:
> - data_row.append(pps[method])
> - data_row.append(pct[method])
> -
> - # generate report table
> - self.result_table_add(data_row)
> - self.l3fwd_test_results['data'].append(data_row)
> -
> def set_up(self):
> """
> Run before each test case.
> """
> pass
>
> - def no_test_perf_l3fwd_4ports(self):
> - """
> - L3fwd main 4 ports.
> - """
> -
> - # Based on h/w type, choose how many ports to use
> - ports = self.dut.get_ports()
> - # Verify that enough ports are available
> - self.verify(len(ports) >= 4, "Insufficient ports for speed
> testing")
> -
> - header_row = ["Frame size", "mode", "S/C/T", "RX Queues/NIC Port"]
> -
> - for method in TestL3fwd.methods:
> - header_row.append('%s Mpps' % method)
> - header_row.append('% linerate')
> -
> - self.result_table_create(header_row)
> - self.l3fwd_test_results['header'] = header_row
> - self.l3fwd_test_results['data'] = []
> -
> - for frame_size in TestL3fwd.frame_sizes:
> -
> - # Prepare traffic flow
> - payload_size = frame_size - \
> - HEADER_SIZE['ip'] - HEADER_SIZE['eth']
> -
> - for _port in range(4):
> - dmac = self.dut.get_mac_address(valports[_port])
> - flows = ['Ether(dst="%s")/%s/("X"*%d)' % (dmac, flow,
> payload_size) for flow in self.flows()[_port * 2:(_port + 1) * 2]]
> - self.tester.scapy_append('wrpcap("dst%d.pcap", [%s])' %
> (valports[_port], string.join(flows, ',')))
> -
> - self.tester.scapy_execute()
> -
> - self.rst_report("Flows for 4 ports, %d frame size.\n" %
> (frame_size),
> - annex=True)
> - self.rst_report("%s" % string.join(flows, '\n'),
> - frame=True, annex=True)
> -
> - # Get the number of sockets of the board
> - number_sockets = self.dut.send_expect("grep
> \"processor\|physical id\|core id\|^$\" /proc/cpuinfo | grep physical
> | sort -u | wc -l", "# ")
> - number_sockets = int(number_sockets.split('\r\n')[0])
> -
> - # Run case by case
> - for test_case in TestL3fwd.test_cases_4_ports:
> -
> - # Check if the board has sockets enough for the test case
> - if number_sockets >= int(test_case[1].split('/')[0][0]):
> - self.get_throughput(frame_size, *test_case)
> -
> - self.result_table_print()
> -
> - def no_test_perf_l3fwd_2ports(self):
> - """
> - L3fwd main 2 ports.
> - """
> -
> - header_row = ["Frame", "mode", "S/C/T", "Mpps", "% linerate",
> "latency_max(us)", "latency_min(us)", "latency_avg(us)"]
> - self.l3fwd_test_results['header'] = header_row
> - self.result_table_create(header_row)
> - self.l3fwd_test_results['data'] = []
> -
> - for frame_size in TestL3fwd.frame_sizes:
> -
> - # Prepare traffic flow
> - payload_size = frame_size - \
> - HEADER_SIZE['ip'] - HEADER_SIZE['eth']
> - for _port in range(2):
> - dmac = self.dut.get_mac_address(valports[_port])
> - flows = ['Ether(dst="%s")/%s/("X"*%d)' % (dmac, flow,
> payload_size) for flow in self.flows()[_port *2:(_port +1)*2]]
> - self.tester.scapy_append('wrpcap("dst%d.pcap",
> [%s])' %(valports[_port],string.join(flows,',')))
> - self.tester.scapy_execute()
> -
> - self.rst_report("Flows for 2 ports, %d frame size.\n" %
> (frame_size),
> - annex=True)
> - self.rst_report("%s" % string.join(flows, '\n'),
> - frame=True, annex=True)
> -
> -
> - # Prepare the command line
> - global corelist
> - pat = re.compile("P([0123]),([0123]),(C\{\d.\d.\d\})")
> -
> - pat2 = re.compile("C\{\d")
> - repl1 = "C{" + str(self.port_socket)
> -
> - coreMask = {}
> - cmdlist = []
> - rtCmdLines = {}
> - cmdlist = TestL3fwd.test_cases_2_ports
> - for cmdline_pat in cmdlist:
> - corelist = []
> - rtCmdLines[cmdline_pat[1]] = cmdline_pat[2]
> - while pat.search(rtCmdLines[cmdline_pat[1]]):
> - # Change the socket to the NIC's socket
> - if cmdline_pat[1].find('1S')>=0:
> - rtCmdLines[cmdline_pat[1]] = pat2.sub(repl1,
> rtCmdLines[cmdline_pat[1]])
> - rtCmdLines[cmdline_pat[1]] = pat.sub(self.repl,
> rtCmdLines[cmdline_pat[1]])
> -
> - self.logger.info("%s\n" % str(corelist))
> - coreMask[cmdline_pat[1]] =
> utils.create_mask(set(corelist))
> -
> - # measure by two different mode
> - for mode in TestL3fwd.methods:
> -
> - # start l3fwd
> - index = 0
> - subtitle = []
> - for cores in rtCmdLines.keys():
> -
> - info = "Executing l3fwd using %s mode, 2 ports, %s
> and %d frame size.\n" % (
> - mode, cores, frame_size)
> -
> - self.logger.info(info)
> - self.rst_report(info, annex=True)
> -
> - subtitle.append(cores)
> - cmdline = rtCmdLines[cores] % (TestL3fwd.path +
> "l3fwd_" + mode, coreMask[cores],
> -
> self.dut.get_memory_channels(), utils.create_mask(valports[:2]))
> -
> - if frame_size > 1518:
> - cmdline = cmdline + " --enable-jumbo --max-pkt-
> len %d" % frame_size
> - self.rst_report(cmdline + "\n", frame=True,
> annex=True)
> -
> - out = self.dut.send_expect(cmdline, "L3FWD:", 120)
> -
> - # Measure test
> - tgenInput = []
> - for rxPort in range(2):
> - # No use on rx/tx limitation
> - if rxPort % 2 == 0:
> - txIntf =
> self.tester.get_local_port(valports[rxPort + 1])
> - else:
> - txIntf =
> self.tester.get_local_port(valports[rxPort - 1])
> -
> - rxIntf =
> self.tester.get_local_port(valports[rxPort])
> - if rxPort % 2 == 0:
> - tgenInput.append((txIntf, rxIntf,
> "dst%d.pcap" %valports[rxPort+1]))
> - else:
> - tgenInput.append((txIntf, rxIntf,
> "dst%d.pcap" %valports[rxPort-1]))
> -
> - _, pps =
> self.tester.traffic_generator_throughput(tgenInput)
> - self.verify(pps > 0, "No traffic detected")
> - pps /= 1000000.0
> - linerate = self.wirespeed(self.nic, frame_size, 2)
> - pct = pps * 100 / linerate
> - latencys =
> self.tester.traffic_generator_latency(tgenInput)
> -
> - index += 1
> -
> - # Stop l3fwd
> - self.dut.send_expect("^C", "#")
> -
> - for latency in latencys:
> - data_row = [frame_size, mode, cores, str(pps),
> str(pct), str(latency['max']), str(latency['min']),
> str(latency['average'])]
> - self.result_table_add(data_row)
> - self.l3fwd_test_results['data'].append(data_row)
> -
> - self.result_table_print()
> -
> def test_perf_rfc2544(self):
>
> ports = self.dut.get_ports()
> @@ -480,10 +245,11 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
>
> # Prepare traffic flow
> payload_size = frame_size - \
> - HEADER_SIZE['ip'] - HEADER_SIZE['eth']
> + HEADER_SIZE['ip'] - HEADER_SIZE['eth'] -
> HEADER_SIZE['tcp']
> +
> for _port in range(ports_num):
> dmac = self.dut.get_mac_address(valports[_port])
> - flows = ['Ether(dst="%s")/%s/("X"*%d)' % (dmac, flow,
> payload_size) for flow in self.flows()[_port *2:(_port +1)*2]]
> + flows = ['Ether(dst="%s")/%s/TCP()/("X"*%d)' % (dmac,
> flow, payload_size) for flow in self.flows()[_port *2:(_port +1)*2]]
> self.tester.scapy_append('wrpcap("dst%d.pcap",
> [%s])' %(valports[_port],string.join(flows,',')))
> self.tester.scapy_execute()
>
> @@ -543,6 +309,8 @@ class TestL3fwd(TestCase,IxiaPacketGenerator):
> subtitle.append(cores)
> cmdline = rtCmdLines[cores] % (TestL3fwd.path
> + "l3fwd_" + mode, coreMask[cores],
>
> self.dut.get_memory_channels(),
> utils.create_mask(valports[:ports_num]))
> + if self.nic == "niantic":
> + cmdline = cmdline + " --parse-ptype"
>
Enabled software parse-type will harm the performance, is it necessary?
For Niantic, run l3fwd, need this param, otherwise it will fail. And seems that do not affect Niantic performance data.
Best Regards
Yulong Pei
> if frame_size > 1518:
> cmdline = cmdline + " --enable-jumbo
> --max- pkt-len %d" % frame_size @@ -571,9 +339,10 @@ class
> TestL3fwd(TestCase,IxiaPacketGenerator):
> linerate = self.wirespeed(self.nic,
> frame_size,
> ports_num)
> zero_loss_throughput = (linerate *
> zero_loss_rate) / 100
>
> - tx_pkts = human_read_number(tx_pkts)
> - rx_pkts = human_read_number(rx_pkts)
> - loss_pkts = human_read_number(loss_pkts)
> + tx_pkts = utils.human_read_number(tx_pkts)
> + rx_pkts = utils.human_read_number(rx_pkts)
> + loss_pkts =
> + utils.human_read_number(loss_pkts)
> +
>
> #data_row = [frame_size, mode, cores,
> tx_pkts, rx_pkts, loss_pkts, zero_loss_rate, zero_loss_throughput]
> data_row = [frame_size, mode, cores,
> zero_loss_throughput, zero_loss_rate]
> --
> 2.1.0
next prev parent reply other threads:[~2017-03-02 2:49 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <1488418539-8237-1-git-send-email-yulong.pei@intel.com>
2017-03-02 1:38 ` [dts] [PATCH 1/2] tests pmd: do changes due to " Liu, Yong
2017-03-02 1:52 ` Pei, Yulong
[not found] ` <1488418539-8237-2-git-send-email-yulong.pei@intel.com>
2017-03-02 1:53 ` [dts] [PATCH 2/2] tests l3fwd: fix issue for " Liu, Yong
2017-03-02 2:49 ` Pei, Yulong [this message]
2017-03-02 3:10 ` Liu, Yong
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=188971FCDA171749BED5DA74ABF3E6F03B690748@shsmsx102.ccr.corp.intel.com \
--to=yulong.pei@intel.com \
--cc=dts@dpdk.org \
--cc=yong.liu@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).