AnyEvent-FastPing/FPing.pm

=head1 NAME

Net::FPing - quickly ping a large number of hosts

=head1 SYNOPSIS

 use Net::FPing;

=head1 DESCRIPTION

This module was written for a signle purpose only: sendinf ICMP EHCO
REQUEST packets as quickly as possible to a large number of hosts
(thousands to millions).

It employs a sending thread and is fully event-driven (using AnyEvent), so
you have to run an event model supported by AnyEvent to use this module.

Receiving and processing reply packets is not currently supported by this
module.

=over 4

=cut

package Net::FPing;

use AnyEvent;

BEGIN {
   $VERSION = '0.01';
   @ISA = qw(Exporter);

   require Exporter;
   Exporter::export_ok_tags(keys %EXPORT_TAGS);

   require XSLoader;
   XSLoader::load __PACKAGE__, $VERSION;
}

our $THR_REQ_FH; open $THR_REQ_FH, ">&=$THR_REQ_FD" or die "FATAL: cannot fdopen";
our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen";

our $THR_REQ_W;
our $THR_RES_W = AnyEvent->io (fh => $THR_RES_FH, poll => 'r', cb => sub {
   my $sv = _read_res
      or return;

   $sv->();
});

our @THR_REQ_BUF;

sub _send_req($) {
   $THR_REQ_BUF .= $_[0];

   $THR_REQ_W ||= AnyEvent->io (fh => $THR_REQ_FH, poll => 'w', cb => sub {
      my $len = syswrite $THR_REQ_FH, $THR_REQ_BUF;
      substr $THR_REQ_BUF, 0, $len, "";

      undef $THR_REQ_W unless length $THR_REQ_BUF;
   });
}

=item Net::FPing::ipv4_supported

Returns true if IPv4 is supported in this module and on this system.

=item Net::FPing::ipv6_supported

Returns true if IPv6 is supported in this module and on this system.

=item Net::FPing::ipv4_pktsize

Returns the number of bytes each IPv4 ping packet has.

=item Net::FPing::ipv6_pktsize

Returns the number of bytes each IPv4 ping packet has.

=item Net::FPing::icmp_ping [ranges...], $send_interval, $payload, \&callback

Ping the given IPv4 address ranges. Each range is an arrayref of the
form C<[lo, hi, interval]>, where C<lo> and C<hi> are octet strings with
either 4 octets (for IPv4 addresses) or 16 octets (for IPV6 addresses),
representing the lowest and highest address to ping (you can convert a
dotted-quad IPv4 address to this format by using C<inet_aton $address>. The
range C<interval> is the minimum time in seconds between pings to the
given range. If omitted, defaults to C<$send_interval>.

The C<$send_interval> is the minimum interval between sending any two
packets and is a way to make an overall rate limit. If omitted, pings will
be send as fast as possible.

The C<$payload> is a 32 bit integer given as the ICMP ECHO REQUEST ident
and sequence numbers (in unspecified order :).

The request will be queued and all requests will be served by a background
thread in order. When all ranges have been pinged, the C<callback> will be
called.

Algorithm: Each range has an associated "next time to send packet"
time. The algorithm loops as long as there are ranges with hosts to be
pinged and always serves the range with the most urgent packet send
time. It will at most send one packet every C<$send_interval> seconds. The
algorithm to send each packet is O(log n) on the number of ranges, so even
a large number of ranges (many thousands) is managable. No storage is
allocated per address.

Performance: On my 2 GHZ Opteron system with a pretty average nvidia
gigabit network card I can ping around 60k to 200k adresses per second,
depending on routing decisions.

Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and
11.0.0.1-11.0.255.255 with at most 1000 packets/s. Do not, however, exceed
1000 packets/s overall:

   Net::FPing::icmp_ping
      [v10.0.0.1, v10.0.0.15, .01],
      [v11.0.0.1, v11.0.255.255, .001],
      .001, 0x12345678,
      sub {
         warn "all ranges pinged\n";
      }
   ;

=cut

sub icmp_ping($$$&) {
   _send_req _req_icmp_ping @_;
}

our $ICMP4_FH;
our $ICMP4_W = (open $ICMP4_FH, "<&=$ICMP4_FD") && AnyEvent->io (fh => $ICMP4_FH, poll => 'r', cb => \&_recv_icmp4);
our $ICMP6_FH;
our $ICMP6_W = (open $ICMP6_FH, "<&=$ICMP6_FD") && AnyEvent->io (fh => $ICMP6_FH, poll => 'r', cb => \&_recv_icmp6);

1;

=back

=head1 BUGS AND SHORTCOMINGS

   - replies are not yet accessible.
   - ipv6 support has never ever been tested.

=head1 AUTHOR

 Marc Lehmann <schmorp@schmorp.de>
 http://home.schmorp.de/

=head1 AUTHOR

 This software is distributed under the GENERAL PUBLIC LICENSE,
 version 2 or any later.

=cut

Revision:	1.2
Committed:	Fri May 4 07:36:45 2007 UTC (18 years, 9 months ago) by root
Branch:	MAIN
Changes since 1.1:	+28 -14 lines
Log Message:	* empty log message *
#	User	Rev	Content
1	root	1.1	=head1 NAME
2
3			Net::FPing - quickly ping a large number of hosts
4
5			=head1 SYNOPSIS
6
7			use Net::FPing;
8
9			=head1 DESCRIPTION
10
11			This module was written for a signle purpose only: sendinf ICMP EHCO
12			REQUEST packets as quickly as possible to a large number of hosts
13			(thousands to millions).
14
15			It employs a sending thread and is fully event-driven (using AnyEvent), so
16			you have to run an event model supported by AnyEvent to use this module.
17
18			Receiving and processing reply packets is not currently supported by this
19			module.
20
21			=over 4
22
23			=cut
24
25			package Net::FPing;
26
27			use AnyEvent;
28
29			BEGIN {
30			$VERSION = '0.01';
31			@ISA = qw(Exporter);
32
33			require Exporter;
34			Exporter::export_ok_tags(keys %EXPORT_TAGS);
35
36			require XSLoader;
37			XSLoader::load __PACKAGE__, $VERSION;
38			}
39
40			our $THR_REQ_FH; open $THR_REQ_FH, ">&=$THR_REQ_FD" or die "FATAL: cannot fdopen";
41			our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen";
42
43			our $THR_REQ_W;
44			our $THR_RES_W = AnyEvent->io (fh => $THR_RES_FH, poll => 'r', cb => sub {
45			my $sv = _read_res
46			or return;
47
48			$sv->();
49			});
50
51			our @THR_REQ_BUF;
52
53			sub _send_req($) {
54			$THR_REQ_BUF .= $_[0];
55
56			$THR_REQ_W \|\|= AnyEvent->io (fh => $THR_REQ_FH, poll => 'w', cb => sub {
57			my $len = syswrite $THR_REQ_FH, $THR_REQ_BUF;
58			substr $THR_REQ_BUF, 0, $len, "";
59
60			undef $THR_REQ_W unless length $THR_REQ_BUF;
61			});
62			}
63
64	root	1.2	=item Net::FPing::ipv4_supported
65
66			Returns true if IPv4 is supported in this module and on this system.
67
68			=item Net::FPing::ipv6_supported
69
70			Returns true if IPv6 is supported in this module and on this system.
71
72			=item Net::FPing::ipv4_pktsize
73
74			Returns the number of bytes each IPv4 ping packet has.
75
76			=item Net::FPing::ipv6_pktsize
77
78			Returns the number of bytes each IPv4 ping packet has.
79
80			=item Net::FPing::icmp_ping [ranges...], $send_interval, $payload, \&callback
81	root	1.1
82			Ping the given IPv4 address ranges. Each range is an arrayref of the
83			form C<[lo, hi, interval]>, where C<lo> and C<hi> are octet strings with
84			either 4 octets (for IPv4 addresses) or 16 octets (for IPV6 addresses),
85			representing the lowest and highest address to ping (you can convert a
86			dotted-quad IPv4 address to this format by using C<inet_aton $address>. The
87			range C<interval> is the minimum time in seconds between pings to the
88			given range. If omitted, defaults to C<$send_interval>.
89
90			The C<$send_interval> is the minimum interval between sending any two
91			packets and is a way to make an overall rate limit. If omitted, pings will
92			be send as fast as possible.
93
94			The C<$payload> is a 32 bit integer given as the ICMP ECHO REQUEST ident
95			and sequence numbers (in unspecified order :).
96
97			The request will be queued and all requests will be served by a background
98			thread in order. When all ranges have been pinged, the C<callback> will be
99			called.
100
101			Algorithm: Each range has an associated "next time to send packet"
102			time. The algorithm loops as long as there are ranges with hosts to be
103			pinged and always serves the range with the most urgent packet send
104			time. It will at most send one packet every C<$send_interval> seconds. The
105			algorithm to send each packet is O(log n) on the number of ranges, so even
106	root	1.2	a large number of ranges (many thousands) is managable. No storage is
107			allocated per address.
108	root	1.1
109			Performance: On my 2 GHZ Opteron system with a pretty average nvidia
110			gigabit network card I can ping around 60k to 200k adresses per second,
111			depending on routing decisions.
112
113			Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and
114			11.0.0.1-11.0.255.255 with at most 1000 packets/s. Do not, however, exceed
115	root	1.2	1000 packets/s overall:
116	root	1.1
117	root	1.2	Net::FPing::icmp_ping
118	root	1.1	[v10.0.0.1, v10.0.0.15, .01],
119			[v11.0.0.1, v11.0.255.255, .001],
120			.001, 0x12345678,
121			sub {
122			warn "all ranges pinged\n";
123			}
124			;
125
126			=cut
127
128	root	1.2	sub icmp_ping($$$&) {
129			_send_req _req_icmp_ping @_;
130	root	1.1	}
131
132			our $ICMP4_FH;
133			our $ICMP4_W = (open $ICMP4_FH, "<&=$ICMP4_FD") && AnyEvent->io (fh => $ICMP4_FH, poll => 'r', cb => \&_recv_icmp4);
134			our $ICMP6_FH;
135			our $ICMP6_W = (open $ICMP6_FH, "<&=$ICMP6_FD") && AnyEvent->io (fh => $ICMP6_FH, poll => 'r', cb => \&_recv_icmp6);
136
137	root	1.2	1;
138	root	1.1
139	root	1.2	=back
140	root	1.1
141	root	1.2	=head1 BUGS AND SHORTCOMINGS
142	root	1.1
143	root	1.2	- replies are not yet accessible.
144			- ipv6 support has never ever been tested.
145	root	1.1
146			=head1 AUTHOR
147
148			Marc Lehmann <schmorp@schmorp.de>
149			http://home.schmorp.de/
150
151			=head1 AUTHOR
152
153			This software is distributed under the GENERAL PUBLIC LICENSE,
154			version 2 or any later.
155
156			=cut
157