X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=yardstick%2Fnetwork_services%2Ftraffic_profile%2Frfc2544.py;h=e33c437c9d04f8ced99d05174dd842e0446efdaa;hb=b086ecc10ad344b2e8c80be8881cd20ab2c2a7c4;hp=83020c85c1791aa7ca236c7aa54826cd55be2ffd;hpb=9c3db265f442b7db607cb929adf6c89c952add97;p=yardstick.git diff --git a/yardstick/network_services/traffic_profile/rfc2544.py b/yardstick/network_services/traffic_profile/rfc2544.py index 83020c85c..e33c437c9 100644 --- a/yardstick/network_services/traffic_profile/rfc2544.py +++ b/yardstick/network_services/traffic_profile/rfc2544.py @@ -11,190 +11,313 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -""" RFC2544 Throughput implemenation """ -from __future__ import absolute_import -from __future__ import division import logging -from trex_stl_lib.trex_stl_client import STLStream -from trex_stl_lib.trex_stl_streams import STLFlowLatencyStats -from trex_stl_lib.trex_stl_streams import STLTXCont +from trex_stl_lib import api as Pkt +from trex_stl_lib import trex_stl_client +from trex_stl_lib import trex_stl_packet_builder_scapy +from trex_stl_lib import trex_stl_streams + +from yardstick.common import constants +from yardstick.network_services.traffic_profile import trex_traffic_profile -from yardstick.network_services.traffic_profile.trex_traffic_profile \ - import TrexProfile LOGGING = logging.getLogger(__name__) +SRC_PORT = 'sport' +DST_PORT = 'dport' + + +class PortPgIDMap(object): + """Port and pg_id mapping class + + "pg_id" is the identification STL library gives to each stream. In the + RFC2544Profile class, the traffic has a STLProfile per port, which contains + one or several streams, one per packet size defined in the IMIX test case + description. + + Example of port <-> pg_id map: + self._port_pg_id_map = { + 0: [1, 2, 3, 4], + 1: [5, 6, 7, 8] + } + """ + + def __init__(self): + self._pg_id = 0 + self._last_port = None + self._port_pg_id_map = {} + + def add_port(self, port): + self._last_port = port + self._port_pg_id_map[port] = [] + + def get_pg_ids(self, port): + return self._port_pg_id_map.get(port, []) + + def increase_pg_id(self, port=None): + port = self._last_port if not port else port + if port is None: + return + pg_id_list = self._port_pg_id_map.get(port) + if not pg_id_list: + self.add_port(port) + pg_id_list = self._port_pg_id_map[port] + self._pg_id += 1 + pg_id_list.append(self._pg_id) + return self._pg_id -class RFC2544Profile(TrexProfile): - """ This class handles rfc2544 implemenation. """ +class RFC2544Profile(trex_traffic_profile.TrexProfile): + """TRex RFC2544 traffic profile""" + + TOLERANCE_LIMIT = 0.01 def __init__(self, traffic_generator): super(RFC2544Profile, self).__init__(traffic_generator) self.generator = None - self.max_rate = None - self.min_rate = None - self.ports = None - self.rate = 100 - self.drop_percent_at_max_tx = None - self.throughput_max = None + self.rate = self.config.frame_rate + self.max_rate = self.config.frame_rate + self.min_rate = 0 + self.drop_percent_max = 0 def register_generator(self, generator): self.generator = generator - def execute_traffic(self, traffic_generator=None): - """ Generate the stream and run traffic on the given ports """ + def stop_traffic(self, traffic_generator=None): + """"Stop traffic injection, reset counters and remove streams""" if traffic_generator is not None and self.generator is None: self.generator = traffic_generator - if self.ports is not None: - return + self.generator.client.stop() + self.generator.client.reset() + self.generator.client.remove_all_streams() - self.ports = [] + def execute_traffic(self, traffic_generator=None): + """Generate the stream and run traffic on the given ports + + :param traffic_generator: (TrexTrafficGenRFC) traffic generator + :return ports: (list of int) indexes of ports + port_pg_id: (dict) port indexes and pg_id [1] map + [1] https://trex-tgn.cisco.com/trex/doc/cp_stl_docs/api/ + profile_code.html#stlstream-modes + """ + if traffic_generator is not None and self.generator is None: + self.generator = traffic_generator + + port_pg_id = PortPgIDMap() + ports = [] for vld_id, intfs in sorted(self.generator.networks.items()): profile_data = self.params.get(vld_id) - # no profile for this port if not profile_data: continue - # correlated traffic doesn't use public traffic? - if vld_id.startswith(self.DOWNLINK) and \ - self.generator.rfc2544_helper.correlated_traffic: + if (vld_id.startswith(self.DOWNLINK) and + self.generator.rfc2544_helper.correlated_traffic): continue for intf in intfs: - port = self.generator.port_num(intf) - self.ports.append(port) - self.generator.client.add_streams(self.get_streams(profile_data), ports=port) - - self.max_rate = self.rate - self.min_rate = 0 - self.generator.client.start(ports=self.ports, mult=self.get_multiplier(), - duration=30, force=True) - self.drop_percent_at_max_tx = 0 - self.throughput_max = 0 - - def get_multiplier(self): - """ Get the rate at which next iteration to run """ - self.rate = round((self.max_rate + self.min_rate) / 2.0, 2) - multiplier = round(self.rate / self.pps, 2) - return str(multiplier) - - def get_drop_percentage(self, generator=None): - """ Calculate the drop percentage and run the traffic """ - if generator is None: - generator = self.generator - run_duration = self.generator.RUN_DURATION - samples = self.generator.generate_samples(self.ports) - - in_packets = sum([value['in_packets'] for value in samples.values()]) - out_packets = sum([value['out_packets'] for value in samples.values()]) - - packet_drop = abs(out_packets - in_packets) + port_num = int(self.generator.port_num(intf)) + ports.append(port_num) + port_pg_id.add_port(port_num) + profile = self._create_profile(profile_data, + self.rate, port_pg_id, + self.config.enable_latency) + self.generator.client.add_streams(profile, ports=[port_num]) + + self.generator.client.start(ports=ports, + duration=self.config.duration, + force=True) + return ports, port_pg_id + + def _create_profile(self, profile_data, rate, port_pg_id, enable_latency): + """Create a STL profile (list of streams) for a port""" + streams = [] + for packet_name in profile_data: + imix = (profile_data[packet_name]. + get('outer_l2', {}).get('framesize')) + imix_data = self._create_imix_data(imix) + self._create_vm(profile_data[packet_name]) + _streams = self._create_streams(imix_data, rate, port_pg_id, + enable_latency) + streams.extend(_streams) + return trex_stl_streams.STLProfile(streams) + + def _create_imix_data(self, imix, + weight_mode=constants.DISTRIBUTION_IN_PACKETS): + """Generate the IMIX distribution for a STL profile + + The input information is the framesize dictionary in a test case + traffic profile definition. E.g.: + downlink_0: + ipv4: + id: 2 + outer_l2: + framesize: + 64B: 10 + 128B: 20 + ... + + This function normalizes the sum of framesize weights to 100 and + returns a dictionary of frame sizes in bytes and weight in percentage. + E.g.: + imix_count = {64: 25, 128: 75} + + The weight mode is described in [1]. There are two ways to describe the + weight of the packets: + - Distribution in packets: the weight defines the percentage of + packets sent per packet size. IXIA uses this definition. + - Distribution in bytes: the weight defines the percentage of bytes + sent per packet size. + + Packet size # packets D. in packets Bytes D. in bytes + 40 7 58.33% 280 7% + 576 4 33.33% 2304 56% + 1500 1 8.33% 1500 37% + + [1] https://en.wikipedia.org/wiki/Internet_Mix + + :param imix: (dict) IMIX size and weight + """ + imix_count = {} + if not imix: + return imix_count + + imix_count = {size.upper().replace('B', ''): int(weight) + for size, weight in imix.items()} + imix_sum = sum(imix_count.values()) + if imix_sum <= 0: + imix_count = {64: 100} + imix_sum = 100 + + weight_normalize = float(imix_sum) / 100 + imix_dip = {size: float(weight) / weight_normalize + for size, weight in imix_count.items()} + + if weight_mode == constants.DISTRIBUTION_IN_BYTES: + return imix_dip + + byte_total = sum([int(size) * weight + for size, weight in imix_dip.items()]) + return {size: (int(size) * weight * 100) / byte_total + for size, weight in imix_dip.items()} + + def _create_vm(self, packet_definition): + """Create the STL Raw instructions""" + self.ether_packet = Pkt.Ether() + self.ip_packet = Pkt.IP() + self.ip6_packet = None + self.udp_packet = Pkt.UDP() + self.udp[DST_PORT] = 'UDP.dport' + self.udp[SRC_PORT] = 'UDP.sport' + self.qinq = False + self.vm_flow_vars = [] + outer_l2 = packet_definition.get('outer_l2') + outer_l3v4 = packet_definition.get('outer_l3v4') + outer_l3v6 = packet_definition.get('outer_l3v6') + outer_l4 = packet_definition.get('outer_l4') + if outer_l2: + self._set_outer_l2_fields(outer_l2) + if outer_l3v4: + self._set_outer_l3v4_fields(outer_l3v4) + if outer_l3v6: + self._set_outer_l3v6_fields(outer_l3v6) + if outer_l4: + self._set_outer_l4_fields(outer_l4) + self.trex_vm = trex_stl_packet_builder_scapy.STLScVmRaw( + self.vm_flow_vars) + + def _create_single_packet(self, size=64): + size -= 4 + ether_packet = self.ether_packet + ip_packet = self.ip6_packet if self.ip6_packet else self.ip_packet + udp_packet = self.udp_packet + if self.qinq: + qinq_packet = self.qinq_packet + base_pkt = ether_packet / qinq_packet / ip_packet / udp_packet + else: + base_pkt = ether_packet / ip_packet / udp_packet + pad = max(0, size - len(base_pkt)) * 'x' + return trex_stl_packet_builder_scapy.STLPktBuilder( + pkt=base_pkt / pad, vm=self.trex_vm) + + def _create_streams(self, imix_data, rate, port_pg_id, enable_latency): + """Create a list of streams per packet size + + The STL TX mode speed of the generated streams will depend on the frame + weight and the frame rate. Both the frame weight and the total frame + rate are normalized to 100. The STL TX mode speed, defined in + percentage, is the combitation of both percentages. E.g.: + frame weight = 100 + rate = 90 + --> STLTXmode percentage = 10 (%) + + frame weight = 80 + rate = 50 + --> STLTXmode percentage = 40 (%) + + :param imix_data: (dict) IMIX size and weight + :param rate: (float) normalized [0..100] total weight + :param pg_id: (PortPgIDMap) port / pg_id (list) map + """ + streams = [] + for size, weight in ((int(size), float(weight)) for (size, weight) + in imix_data.items() if float(weight) > 0): + packet = self._create_single_packet(size) + pg_id = port_pg_id.increase_pg_id() + stl_flow = (trex_stl_streams.STLFlowLatencyStats(pg_id=pg_id) if + enable_latency else None) + mode = trex_stl_streams.STLTXCont(percentage=weight * rate / 100) + streams.append(trex_stl_client.STLStream( + packet=packet, flow_stats=stl_flow, mode=mode)) + return streams + + def get_drop_percentage(self, samples, tol_low, tol_high, + correlated_traffic): + """Calculate the drop percentage and run the traffic""" + completed = False + out_pkt_end = sum(port['out_packets'] for port in samples[-1].values()) + in_pkt_end = sum(port['in_packets'] for port in samples[-1].values()) + out_pkt_ini = sum(port['out_packets'] for port in samples[0].values()) + in_pkt_ini = sum(port['in_packets'] for port in samples[0].values()) + time_diff = (list(samples[-1].values())[0]['timestamp'] - + list(samples[0].values())[0]['timestamp']).total_seconds() + out_packets = out_pkt_end - out_pkt_ini + in_packets = in_pkt_end - in_pkt_ini + tx_rate_fps = float(out_packets) / time_diff + rx_rate_fps = float(in_packets) / time_diff drop_percent = 100.0 - try: - drop_percent = round((packet_drop / float(out_packets)) * 100, 5) - except ZeroDivisionError: - LOGGING.info('No traffic is flowing') - - # TODO(esm): RFC2544 doesn't tolerate packet loss, why do we? - tolerance_low = generator.rfc2544_helper.tolerance_low - tolerance_high = generator.rfc2544_helper.tolerance_high - - tx_rate = out_packets / run_duration - rx_rate = in_packets / run_duration - - throughput_max = self.throughput_max - drop_percent_at_max_tx = self.drop_percent_at_max_tx - if self.drop_percent_at_max_tx is None: - self.rate = tx_rate - self.first_run = False + # https://tools.ietf.org/html/rfc2544#section-26.3 + if out_packets: + drop_percent = round( + (float(abs(out_packets - in_packets)) / out_packets) * 100, 5) - if drop_percent > tolerance_high: - # TODO(esm): why don't we discard results that are out of tolerance? + tol_high = max(tol_high, self.TOLERANCE_LIMIT) + tol_low = min(tol_low, self.TOLERANCE_LIMIT) + if drop_percent > tol_high: self.max_rate = self.rate - if throughput_max == 0: - throughput_max = rx_rate - drop_percent_at_max_tx = drop_percent - - elif drop_percent >= tolerance_low: - # TODO(esm): why do we update the samples dict in this case - # and not update our tracking values? - throughput_max = rx_rate - drop_percent_at_max_tx = drop_percent - - elif drop_percent >= self.drop_percent_at_max_tx: - # TODO(esm): why don't we discard results that are out of tolerance? + elif drop_percent < tol_low: self.min_rate = self.rate - self.drop_percent_at_max_tx = drop_percent_at_max_tx = drop_percent - self.throughput_max = throughput_max = rx_rate - else: - # TODO(esm): why don't we discard results that are out of tolerance? - self.min_rate = self.rate - - generator.clear_client_stats(self.ports) - generator.start_client(self.ports, mult=self.get_multiplier(), - duration=run_duration, force=True) + completed = True - # if correlated traffic update the Throughput - if generator.rfc2544_helper.correlated_traffic: - throughput_max *= 2 - - samples.update({ - 'TxThroughput': tx_rate, - 'RxThroughput': rx_rate, - 'CurrentDropPercentage': drop_percent, - 'Throughput': throughput_max, - 'DropPercentage': drop_percent_at_max_tx, - }) + last_rate = self.rate + self.rate = round(float(self.max_rate + self.min_rate) / 2.0, 5) - return samples + throughput = rx_rate_fps * 2 if correlated_traffic else rx_rate_fps - def execute_latency(self, generator=None, samples=None): - if generator is not None and self.generator is None: - self.generator = generator + if drop_percent > self.drop_percent_max: + self.drop_percent_max = drop_percent - if samples is None: - samples = self.generator.generate_samples() + latency = {port_num: value['latency'] + for port_num, value in samples[-1].items()} - self.pps, multiplier = self.calculate_pps(samples) - self.ports = [] - self.pg_id = self.params['traffic_profile'].get('pg_id', 1) - for vld_id, intfs in sorted(self.generator.networks.items()): - profile_data = self.params.get(vld_id) - if not profile_data: - continue - # correlated traffic doesn't use public traffic? - if vld_id.startswith(self.DOWNLINK) and \ - self.generator.rfc2544_helper.correlated_traffic: - continue - for intf in intfs: - port = self.generator.port_num(intf) - self.ports.append(port) - self.generator.client.add_streams(self.get_streams(profile_data), ports=port) - - self.generator.start_client(ports=self.ports, mult=str(multiplier), - duration=120, force=True) - self.first_run = False - - def calculate_pps(self, samples): - pps = round(samples['Throughput'] / 2, 2) - multiplier = round(self.rate / self.pps, 2) - return pps, multiplier - - def create_single_stream(self, packet_size, pps, isg=0): - packet = self._create_single_packet(packet_size) - if pps: - stl_mode = STLTXCont(pps=pps) - else: - stl_mode = STLTXCont(pps=self.pps) - if self.pg_id: - LOGGING.debug("pg_id: %s", self.pg_id) - stl_flow_stats = STLFlowLatencyStats(pg_id=self.pg_id) - stream = STLStream(isg=isg, packet=packet, mode=stl_mode, - flow_stats=stl_flow_stats) - self.pg_id += 1 - else: - stream = STLStream(isg=isg, packet=packet, mode=stl_mode) - return stream + output = { + 'TxThroughput': tx_rate_fps, + 'RxThroughput': rx_rate_fps, + 'CurrentDropPercentage': drop_percent, + 'Throughput': throughput, + 'DropPercentage': self.drop_percent_max, + 'Rate': last_rate, + 'Latency': latency + } + return completed, output