Presentation on theme: "One-Way Ping - Introduction to OWAMP"— Presentation transcript:
1One-Way Ping - Introduction to OWAMP Dr. Quincy Wu, Associate ProfessorGraduate Institute of Communication EngineeringNational Chi Nan University
2Growth of Internet Number of computers attached to the Internet In 1998, the average rate of new computers being added to the Internet reached more than one per secondAnd has acceleratedComputer Networks and Internets, Douglas E. Comer, Pearson Prentice hall, 2004.
3Growth of Internet (cont.) Plotted on a log scaleThe growth appears approximately linearExponential growthThe Internet has been doubling in size every nine to twelve monthsComputer Networks and Internets, Douglas E. Comer, Pearson Prentice hall, 2004.
5Probing The InternetQ: How do we know the number of computers attached to the Internet?In the early days when the Internet consisted of a dozen sites, this size could be determined manually.Now we use programs that test to see whether a computer is currently online.pingis alivepingis aliveCertainly, this probing is not very precise, for two reasons.
6Interpreting A Ping Response C:\>pingPinging cswww.cse.yzu.edu.tw [ ] with 32 bytes of data:Reply from : bytes=32 time=14ms TTL=115Reply from : bytes=32 time=11ms TTL=115Reply from : bytes=32 time=10ms TTL=115Ping statistics for : Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),Approximate round trip times in milli-seconds: Minimum = 10ms, Maximum = 14ms, Average = 11msC:\>pingPinging [ ] with 32 bytes of data:Reply from : bytes=32 time=6ms TTL=56Ping statistics for : Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),Approximate round trip times in milli-seconds: Minimum = 6ms, Maximum = 6ms, Average = 6ms
8Round-Trip Time Client Server 0.000 ms 9.952 ms 1006.122 ms requestreply9.952 msmsrequestreplyms
9Why Didn’t We Measure One-Way Delay? Asynchronous system clocks would make the measurement result confusing.Sender19:20:21Receiver19:20:1919:20:20Delay = -1 sec !
10ICMP Packet FormatRFC 792 – Internet Control Message Protocol
11Why Do We Favor One-Way Delay? The path from a source to a destination may be different than the path from the destination back to the source ("asymmetric paths").Even when the two paths are symmetric, the behavior of applications can be quite different:File transferWeb browsingIPTV
12Why Can We Measure 1-Way Delay Now? Available Time Source:Cesium oscillator: Definition of time (subject to relativistic effects)Rubidium oscillator: found in cell towers, very stableGPS receiver: accuracy circa 10 nsCDMA receiver: accuracy circa 10 μsThe stratum of any NTP-synchronized device is the stratum of the device it is synchronized to, plus 1.GPS receiver: stratum 0Computer connected to it by a serial line: stratum 1Client that gets the time from that computer: stratum 2Stratum 1 Time Servers:
14OWAMP Design Goals One-Way Active Measurement Protocol RFC 4656, September 2006.Wide deployment of “open” servers would allow measurement of one-way delay to become as commonplace as measurement of RTT using ICMP tools such as ping.
15OWAMP Logical Model Session Sender Session Receiver Server OWAMP-TestSession ReceiverServerOWAMP-ControlOWAMP-ControlControl-ClientFetch-Client
16Commonly Implemented Model Session-SenderControl-ClientFetch-ClientOWAMP-TestSession-ReceiverServerOWAMP-Control
17OWAMP-Test Transport Protocol: Sender/Receiver IP and port numbers: UDPSender/Receiver IP and port numbers:Negotiated by OWAMP-Control messageOWAMP-Test does not run on a fixed portTo prevent some devices may assign higher priorities to these measurement packets
18OWAMP-Test Packet Format Sequence: start with 0; incremented by 1Timestamp: RFC1305 formatPadding is random, but users have an option to configure it to consist of all zeros.Minimum data length: 14 octets
19OWAMP Errors Preliminary Findings: Min error estimates look to be in the usec range.Serialization Delay: ~5usec x 2Get Timestamp: ~15usec x 2Additional error is:Time from userland “send” to 1st byte hits the wireTime from kernel has packet to userland “recv” returnsPotentially recv process data processing before calling “recv”
20Internet2 OWAMP deployment 2 overlapping full meshes (IPv4 & IPv6)11 measurement nodes = 220 ongoing testsUDP singletonssingleton: a single observation of one-way delayRate: 10 packets/secondPacket size: 32-byte payloadResults are continuously streamed back to “Measurement Portal” for long-term archive and data dissemination (Near real-time)* These are the values we are currently we are planning on using, but this is not set in stone.
24R&D Issues Design a system to scale (eliminate centralizations) How to discover OWAMP serversDNS SRV,DHCP option,Multicast addressHow to insert On-Demand tests into regularly-scheduled test setBalance centralization and distributed database requirementDynamically allocated AES keyCurrently, the shared secret between sender and receiver is statically assigned
25Security Considerations Protecting Your OWAMP Testing TrafficTo make it impossible for an attacker to tamper with test results.To make it hard for a party in the middle of the network to make results look "better" than they should be.Preventing Third-Party Denial of ServiceCovert Information ChannelsRequirement to Include AES in ImplementationsResource Use LimitationsDisk, Memory, BandwidthUse of Cryptographic Primitives in OWAMPTLSStream-based. Not suitable for OWAMP-Test.DTLSDuplication and reordering information are missingIPSecFew deploymentsSSH 2-4%HTTPS: %IPsec: 0.05%
26HW 3Install OWAMP client/server on your own hosts. Try to test the one-way delay.Your host may possess a public IP address. If this is not the case for IPv4, at least you know how to get a public IPv6 address.Show me your measurement, and the OWAMP packets which you captured.