Skip to content

GitLab

Commit 48a011cc authored by Leigh B. Stoller's avatar Leigh B. Stoller
Browse files

Switch from using pathrate for the BW tests, to using locally modifed 

version of iperf.
parent 9e62b2e4
Hide whitespace changes
Inline Side-by-side
event/linktest/linktest.pl.in
View file @ 48a011cc
......@@ -10,16 +10,17 @@ use Class::Struct;
use POSIX qw(uname);
use IO::Handle;
use English;
use Socket;
#
# Linktest test script. This script is set up to run as root on
# experiment nodes. It is invoked by the Linktest daemon after the
# daemon receives a Linktest "START" event. The script runs tests using
# ping, rude/crude (a real-time packet emitter/collector) and pathrate
# to test all links in the experiment. If the results show a 99% chance
# that the experiment is configured incorrectly, an error is logged to
# the experiment directory in tbdata/linktest. Valid ranges have been
# calibrated ahead of time.
# daemon receives a Linktest "START" event. The script runs tests
# using ping, rude/crude (a real-time packet emitter/collector) and a
# locally hacked iperf to test all links in the experiment. If the
# results show a 99% chance that the experiment is configured
# incorrectly, an error is logged to the experiment directory in
# tbdata/linktest. Valid ranges have been calibrated ahead of time.
#
sub usage()
{
......@@ -31,10 +32,6 @@ sub usage()
}
##############################################################################
# Constants
##############################################################################
......@@ -45,8 +42,7 @@ use constant PATH_NICKNAME => "@CLIENT_VARDIR@/boot/nickname";
use constant PATH_KEYFILE => "@CLIENT_VARDIR@/boot/eventkey";
use constant PATH_RUDE => "/usr/local/bin/rude";
use constant PATH_CRUDE => "/usr/local/bin/crude";
use constant PATH_PATHRATE_SND => "/usr/local/bin/pathrate_snd";
use constant PATH_PATHRATE_RCV => "/usr/local/bin/pathrate_rcv";
use constant PATH_IPERF => "/usr/local/bin/emulab-iperf";
use constant PATH_EMULAB_SYNC => "@CLIENT_BINDIR@/emulab-sync";
use constant PATH_LTEVENT => "@CLIENT_BINDIR@/ltevent";
use constant PATH_TEVC => "@CLIENT_BINDIR@/tevc";
......@@ -56,10 +52,10 @@ use constant EVENTSERVER => "@EVENTSERVER@";
# log files used by tests.
use constant CRUDE_DAT => "/tmp/crude.dat"; # binary data
use constant RUDE_CFG => "/tmp/rude.cfg";
use constant PATHRATE_DAT => "/tmp/pathrate.dat";
use constant IPERF_DAT => "/tmp/iperf.dat";
# pathrate test limits.
use constant LIMIT_BW_HI => 45000000;
# iperf test limits.
use constant LIMIT_BW_HI => 100000000;
use constant LIMIT_BW_LO => 1000000;
use constant LIMIT_BW_LOSS => 0;
......@@ -67,7 +63,8 @@ use constant LIMIT_BW_LOSS => 0;
# Make sure the error is a certain significance before we start reporting it.
use constant INSIGNIFICANT_LAT_ERROR_LO => 0.50; # ms
use constant INSIGNIFICANT_LAT_ERROR_HI => 3.00; # ms
use constant INSIGNIFICANT_BW_ERROR => 1.00; # mb
use constant INSIGNIFICANT_BW_ERROR_HI => 0.01; # percent.
use constant INSIGNIFICANT_BW_ERROR_LO => 0.05; # percent.
# latency must be corrected for xmit delay under this speed.
use constant LAT_LOW_BW => 10000000;
......@@ -112,7 +109,9 @@ use constant EXIT_OK => 0;
# struct for representing a link.
struct ( edge => {
src => '$',
srcip => '$',
dst => '$',
dstip => '$',
bw => '$',
delay => '$',
loss => '$'});
......@@ -300,17 +299,17 @@ if(&is_special_node()) {
# run, if any.
#
&do_unlink(CRUDE_DAT);
&do_unlink(PATHRATE_DAT);
&do_unlink(IPERF_DAT);
&do_unlink(RUDE_CFG);
#
# Start up child processes for receiving RUDE and Pathrate streams.
# Start up child processes for receiving RUDE and iperf streams.
# These run over the lifetime of Linktest to reduce the number of
# barrier synchronizations and startup delays. Always give the
# collectors a moment to start up.
#
if(&dotest(TEST_BW)){
&my_system_initonly(PATH_PATHRATE_SND,"-i","-q");
&my_system_initonly(PATH_IPERF,"-s","-f","b","-u","-w","200000");
sleep(1);
}
if(&dotest(TEST_LOSS)){
......@@ -748,71 +747,124 @@ sub valid_bw {
# Checks bandwidth for directly connected links.
sub bw_test {
my @analyze_list;
my @edge_copy;
@edge_copy = @links;
my @analyze_list = ();
my @edge_copy = @links;
#
# all nodes will execute the same reductions on the edge list
# on their own so that the number of barriers is the same.
while(&has_elems(\@edge_copy)) {
my $edge = &get_assign(\@edge_copy);
#
while (&has_elems(\@edge_copy)) {
my ($edge,$redge) = &get_twoway_assign(\@edge_copy);
if(defined($edge) ) {
if (defined($edge)) {
if($hostname eq $edge->dst) {
#
# iperf does a twoway test.
#
if (&valid_bw($edge)) {
push @analyze_list, $edge;
push(@analyze_list, $edge);
&info(" Starting bandwidth test for " .
&print_link($edge) . "\n");
}
else {
&debug("Skipping bandwidth test for " .
&print_link($edge) . "\n");
&my_system(PATH_PATHRATE_RCV, "-Q","-s",$edge->src,"-q","-N", PATHRATE_DAT);
} else {
&debug("Skipping bandwidth test for " . &print_link($edge) . "\n");
&info("*** Skipping bandwidth test for " .
&print_link($edge) . "\n");
&info("*** Bandwidth is out of range ".
"(" . LIMIT_BW_LO . " <= BW <= " . LIMIT_BW_HI .") ".
"or loss is too high (> " . LIMIT_BW_LOSS . ").\n");
}
if (&valid_bw($redge)) {
push(@analyze_list, $redge);
&info(" Starting bandwidth test for " .
&print_link($redge) . "\n");
}
else {
&debug("Skipping bandwidth test for " .
&print_link($redge) . "\n");
&info("*** Skipping bandwidth test for " .
&print_link($redge) . "\n");
&info("*** Bandwidth is out of range ".
"(" . LIMIT_BW_LO . " <= BW <= " . LIMIT_BW_HI .") ".
"or loss is too high (> " . LIMIT_BW_LOSS . ").\n");
}
# Okay, start the test.
if (&valid_bw($edge) || &valid_bw($redge)) {
&my_system(PATH_IPERF,
"-c", $edge->src, "-t", "10", "-f", "b",
"-r", "-u", "-w", "200000", "-b", "100Mb",
"-x", "s", "-y", "c",
"-L", "4444", "-o", IPERF_DAT);
}
}
}
&barrier();
}
# read the log file.
if(@analyze_list) {
my @results = &read_file(PATHRATE_DAT);
foreach (@results) {
my $edge = shift(@analyze_list);
my $sender = $edge->src;
if(/SNDR=$sender.*CAPL=(\d+\.\d+)Mbps.*CAPH=(\d+\.\d+)Mbps/) {
my $expected = $edge->bw / 1000000;
my $low = $1;
my $hi = $2;
my $diff;
if($expected > $2) {
$diff = $expected - $2;
} elsif ($expected < $1) {
$diff = $1 - $expected;
} else {
$diff = 0;
if (@analyze_list) {
my @results = &read_file(IPERF_DAT);
foreach my $edge (@analyze_list) {
my $found_results = 0;
foreach my $result (@results) {
my @stuff = split(",", $result);
if (scalar(@stuff) < 9) {
die("Error parsing " . IPERF_DAT . "\n");
}
my $output = sprintf "Receive from " . $edge->src
. ": $1/$2/%.1f/%.1f\n", $expected, $diff;
&debug($output);
if($diff > INSIGNIFICANT_BW_ERROR) {
&error (NAME_BW, $edge, "Bandwidth estimate $low to $hi Mbps does not include expected bandwidth $expected Mbps");
my $myip = $stuff[1];
my $port = $stuff[2];
my $hisip = $stuff[3];
my $bw = $stuff[8];
#
# iperf is a twoway test. Both edges represented in the file.
#
if (($hostname eq $edge->dst &&
$edge->dstip eq $myip && "$port" eq "4444" &&
$edge->srcip eq $hisip) ||
($hostname eq $edge->src &&
$edge->dstip eq $myip && "$port" eq "5001" &&
$edge->srcip eq $hisip)) {
my $expected = $edge->bw;
my $diff = abs($bw - $expected);
my $error = undef;
&debug("BW results for " . $edge->src . ": ".
"$bw/$expected/$diff\n");
#
# The measurement tool does not give perfect results.
# However, it reports low all the time, so if it reports
# high, then the link is almost certainly bad.
#
if ($bw > $edge->bw) {
if ($diff > ($expected * INSIGNIFICANT_BW_ERROR_HI)) {
$error = "higher";
}
}
elsif ($bw < $edge->bw) {
if ($diff > ($expected * INSIGNIFICANT_BW_ERROR_LO)) {
$error = "lower";
}
}
if (defined($error)) {
&error(NAME_BW, $edge,
"Bandwidth estimate ($bw bps) is $error ".
"expected ($expected)");
}
$found_results = 1;
last;
}
} else {
die ("Error while parsing " . PATHRATE_DAT . "\n");
}
if (!$found_results) {
&error(NAME_BW, $edge, "Could not find results!");
}
}
}
......@@ -922,7 +974,9 @@ sub get_topo {
} elsif ( /^l (\S+)\s+(\S+)\s+(\d+)\s+(\d+\.\d+)\s+(\d+\.\d+)/) {
my $newEdge = new edge;
$newEdge->src($1);
$newEdge->srcip(hostip($1));
$newEdge->dst($2);
$newEdge->dstip(hostip($2));
$newEdge->bw($3);
$newEdge->delay($4 * 1000); # units of ms
$newEdge->loss($5);
......@@ -1358,3 +1412,10 @@ sub linux_version {
}
return $vers;
}
sub hostip {
my $host = shift(@_);
my (undef,undef,undef,undef,@ipaddrs) = gethostbyname($host);
return inet_ntoa($ipaddrs[0]);
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment