<?xml version="1.0" standalone="no"?>
                    <!DOCTYPE div SYSTEM "/www/backend/www-xml-443/dtd/caidaML.dtd">
                    <!-- do NOT ERASE the DOCTYPE declaration! --><div>


<tr bgcolor="#f4f4f4">
  <td>
<font face="helvetica,arial" size="2">
<b>URL:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
<a href="ftp://ftp.research.microsoft.com/pub/tr/tr-2000-08.ps">ftp://ftp.research.microsoft.com/pub/tr/tr-2000-08.ps</a><br/>
<a href="http://www.acm.org/sigcomm/sigcomm2000/conf/abstract/5-2.htm">http://www.acm.org/sigcomm/sigcomm2000/conf/abstract/5-2.htm</a>
</font>
  </td>
</tr>


<tr bgcolor="#e9e9e9">
  <td>
<font face="helvetica,arial" size="2">
<b>Entry Date:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
2002-11-4


</font>
  </td>
</tr>


<tr bgcolor="#f4f4f4">
  <td>
<font face="helvetica,arial" size="2">
<b>Abstract:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
This paper examines the latency in Internet path failure, failover and
repair due to the convergence properties of inter-domain
routing. Unlike switches in the public telephony network which exhibit
failover on the order of milliseconds, we show that inter-domain
routers in the packet switched Internet may take several minutes to
reach a consistent view of the network topology after a fault. These
delays stem from temporary routing table oscillations formed during
operation of the BGP path selection process on Internet backbone
routers. During these periods of delayed convergence, end-to-end
Internet paths will experience intermittent loss of connectivity, as
well as increased packet loss and latency. We present a two-year study
of Internet routing convergence through the experimental
instrumentation of key portions of the Internet infrastructure,
including both passive data collection and fault-injection machines at
major Internet exchange points. Based on data from the injection and
measurement of several hundred thousand inter-domain routing faults,
we describe several unexpected properties of convergence and show that
the measured upper bound on Internet inter-domain routing convergence
delay is an order of magnitude slower than previously thought. Our
analysis also shows that the upper computational bound on the number
of router states and control messages exchanged during the process of
BGP convergence is exponential with respect to the number of
autonomous systems on the Internet. Finally, we demonstrate that much
of the observed convergence delay stems from both specific router
vendor implementation decisions, as well as ambiguity in the BGP
specification.


</font>
  </td>
</tr>


<tr bgcolor="#e9e9e9">
  <td>
<font face="helvetica,arial" size="2">
<b>Experiments:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
Over a two year period the authors injected faults at major Internet 
    exchange points, and performed passive measurements.


</font>
  </td>
</tr>


<tr bgcolor="#f4f4f4">
  <td>
<font face="helvetica,arial" size="2">
<b>Results:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
<ul>
    <li> The authors examine the latency in Internet path failure, failover and
      repair due to the convergence properties of inter-domain routing.</li>

    <li> The delay in Internet inter-domain path failovers averages three minutes,
      and some percentage of failovers trigger routing table oscillations
      lasting up to fifteen minutes. The upper bound on delay is an order of
      magnitude slower than previously thought.</li>

    <li>The theoretical upper computational bound on the number of router states
      and control messages during BGP convergence is exponential in the number
      of autonomous systems in the Internet.</li>

    <li>The theoretical lower bound on delay and number of states of BGP
      convergence is linear in the number of autonomous systems in the
      Internet.  The authors suggest changes that would make the time
      complexity a constant (30 seconds). </li>

    <li>Internet path failover affects end-to-end performance - measured packet
      loss grows by a factor of 30 and latency by a factor of four during path
      restoral.</li>

    <li>Much of the observed convergence delay stems from router vendor
      implementation decisions and BGP specification ambiguity.</li>
</ul>


</font>
  </td>
</tr>


<tr bgcolor="#e9e9e9">
  <td>
<font face="helvetica,arial" size="2">
<b>References:</b>
</font>
</td>
  <td>
<font face="helvetica,arial" size="2">
<ul>
  <li>
  Complements:
    <ul>
    <li> T.G. Griffin, F. B. Shepherd, and G. Wilfong, "Policy disputes in path-vector protocols," in Proc. International Conference on Network Protocols, November 1999. </li>
    <li> K. Varadhan, R. Govindan, and D. Estrin. Persistent route oscillations in inter-domain routing. Technical Report 96-631, USC/ISI, February 1996. </li>
    </ul>
  </li>

  <li>
  Explains observations in:
    <ul>
    <li>    C. Labovitz, G. R. Malan, and F. Jahanian, "Origins of pathological Internet routing instability," in Proc. IEEE INFOCOM, March 1999.</li>
    <li>    V. Paxson, "End-to-end Internet packet dynamics," in Proc. ACM SIGCOMM, 1997. </li>
    </ul>
  </li>

  <li>
  Other references:
    <ul>
    <li>    C. Labovitz, A. Ahuja, and F. Jahanian, "Experimental study of Internet stability and wide-area network failures," in Proc. International Symposium on Fault-Tolerant Computing, June 1999.</li>
    <li>    C. Labovitz, G. R. Malan, and F. Jahanian, "Internet routing instability," IEEE/ACM Transactions on Networking, August 1997.</li>
    </ul>
  </li>
</ul>




</font>
  </td>
</tr>
</div>

