Getting Started Guide
High-Performance Network Connections
-
Benefits
to Connecting to a High-Performance Network
-
Learning
About Your Connections with Informational Tools
-
Traceroute
-
Two vBNS Connected Sites
-
vBNS and Abilene Connected
Sites
-
Two Abilene Connected Sites
-
HPN to Commodity Site
-
ping Utility
-
Connecting to an HPN
-
Getting Help Once Connected
1.
Benefits to Connecting to a High-Performance Network
A High-Performance Network (HPN) is a wide-area backbone
service that has high bandwidth, low latency, and small jitter. The vBNS
and Abilene networks (see below) are both classified as HPNs because they
offer much higher quality of network service than the commodity Internet
(the network resources offered by Internet Service Providers). HPNs:
-
give individual users access to all of, or significant fractions of, fast
backbone links that have peak bandwidth at or over 600 Mbits/sec. This
bandwidth may accommodate large dataset applications such as data mining,
scientific simulation, and visualization.
-
have low latency (time between a request and a response)
and delay jitter (variance of delay in packet transmittal). Low
latency is helpful for real-time applications, such as audio and video
conferencing.
-
are continually trying to deploy emerging network technologies,
such as Quality of Service (QoS) features. QoS represents a promise of
network transmission characteristics to an application. The typical QoS
parameters include minimal bandwidth, maximum delay, and jitter.
-
are designed to be highly available and more reliable because
of central management and dedicated resources.
vBNS is the very
high performance backbone network service developed by MCI and NSF. As
of August 1999, 104
institutions were connected to the vBNS network.
Abilene
is HPN service provided by the University Corporation for Advanced Internet
Development (UCAID) that has over 150 member institutions in the
United States. Almost 100
institutions were connected to the Abilene network as of December 1999, and as
of May of 2001 the
number of connected institutions has reached over 200.
Check both lists for the latest figures and
to see if your institution is connected to one of these HPNs.
2.
Learning About Your Connections with Informational Tools
The UNIX traceroute command (tracert in
Windows) is an easy way to check your network connections. Both commands
provide you with information to determine if you are connected to an HPN.
The UNIX ping command returns information about
the round-trip-time from your system to another system and back. The same
command is also available in Windows.
On a UNIX system, enter man commandname to get
more information about a command. In Windows 2000, use the F1
key and search in the help for information about a command.
2.1 Traceroute
The traceroute command (available on UNIX and Windows systems)
is an easy informational tool to use to discover network connectivity.
Entering traceroute without any parameter returns
a help message. On UNIX, enter the command in a shell window. On Windows,
you open a DOS command window and enter the command.
Traceroute uses the ICMP protocol implemented at the routers to process an
ICMP response. ICMP refers to the Internet Control Message Protocol that
handles message control and error-reporting between hosts and routers. High
delays at a hop usually signal that the router is very busy. Not all
applications will experience these delays because routers may treat the
application's IP packets differently. In addition, some routers have buggy ICMP
implementation or simply block the ICMP packets, which would make it appear as if the
router is down or that the remote host is unreachable.
Two vBNS Connected Sites
Below is an example of traceroute from the National
Center for Supercomputing Applications (NCSA) to the San Diego Supercomputing
Center (SDSC). Both of the centers are vBNS connected.
modi4 52% traceroute www.sdsc.edu
traceroute to www.sdsc.edu (198.202.75.101), 30 hops max, 40 byte packets
1 lucy.ncsa.uiuc.edu (141.142.7.228) 1 ms 1 ms 1 ms
2 charlie-atm1-0-8.ncsa.uiuc.edu (141.142.11.134) 1 ms 1 ms 1
ms
3 cs-atm0-0-12.ncsa.vbns.net (141.142.11.1) 1 ms 1 ms 1 ms
4 cs-atm0-0-5.sdsc.vbns.net (204.147.129.69) 55 ms 56 ms 63 ms
5 192.12.207.14 (192.12.207.14) 63 ms 57 ms 56 ms
6 webfarm1.sdsc.edu (198.202.75.201) 63 ms 56 ms 56
ms
The output of traceroute consists of a number of
lines. The first line returned after the ENTER key is pressed tells you
which destination you are tracing to (www.sdsc.edu with an IP address of
198.202.75.101), the maximum number of hops to trace (30), and the size
of the packet (40 bytes, the default). The numbered lines are the responses
from the various routers encountered on the way from Illinois to California.
For each router, traceroute sends three packets and hopefully
gets three responses. The delays of those three packets are printed after
each router name (numbered lines 1 through 6). By looking at the routers
you can see that the traceroute traffic traversed two vBNS routers,
denoted by "vbns.net" in the domain name. This tells you that NCSA is connected
to the vBNS. Try this on a UNIX system at your site and see if "vbns.net"
is returned in your output.
You can specify the packet size on the command line if
you want. The exact syntax varies a little. One example is traceroute
www.sdsc.edu 100 to send out 100-byte packets. But it may send out
140-byte packets on another machine depending on whether traceroute
takes the argument as payload size or the whole ICMP packet size. (Internet
Control Message Protocol is a message control and error-reporting protocol
between a host server and a gateway to the Internet). So you have to experiment
with it and look at the results. On Windows, an example is tracert
www.sdsc.edu. You cannot specify the packet size of tracert
on Windows.
vBNS and Abilene
Connected Sites
The traceroute example below shows a connection
point between a vBNS site (NCSA) and an Abilene site (the University of
Kansas).
modi4 53% traceroute www.ukans.edu
traceroute to raven.cc.ukans.edu (129.237.33.3), 30 hops max, 40 byte packets
1 lucy.ncsa.uiuc.edu (141.142.7.228) 4 ms 4 ms 1 ms
2 charlie-atm1-0-8.ncsa.uiuc.edu (141.142.11.134) 2 ms 10 ms
11 ms
3 cs-atm0-0-12.ncsa.vbns.net (141.142.11.1) 9 ms 10 ms 2 ms
4 cs-atm0-0-6.wae.vbns.net (204.147.129.238) 22 ms 22 ms 21 ms
5 abilene-vbns.abilene.ucaid.edu (198.32.11.9) 32 ms (ttl=250!) 33
ms (ttl=250!) 27 ms (ttl=250!)
6 clev-nycm.abilene.ucaid.edu (198.32.8.29) 30 ms (ttl=251!) 32 ms
(ttl=251!) 31 ms (ttl=251!)
7 ipls-clev.abilene.ucaid.edu (198.32.8.25) 28 ms (ttl=252!) 27 ms
(ttl=252!) 27 ms (ttl=252!)
8 kscy-ipls.abilene.ucaid.edu (198.32.8.5) 37 ms (ttl=251!) 38 ms
(ttl=251!) 37 ms (ttl=251!)
9 ks-2-p00.r.greatplains.net (164.113.238.194) 39 ms (ttl=250!) 37
ms (ttl=250!) 35 ms (ttl=250!)
10 ks-2-ku.r.greatplains.net (164.113.234.205) 37 ms (ttl=249!) 38 ms
(ttl=249!) 39 ms (ttl=249!)
11 129.237.1.254 (129.237.1.254) 43 ms (ttl=248!) 40 ms (ttl=248!)
38 ms (ttl=248!)
12 raven.cc.ukans.edu (129.237.33.3) 39 ms (ttl=56!) 38 ms (ttl=56!)
39 ms (ttl=56!)
The traffic travels through some vBNS routers (vbns.net,
lines 3 and 4) to the Abilene network, which include "abilene.ucaid.edu"
in the domain name (lines 5 through 8). vBNS and Abilene are peered
(joined) together at a few points around the U.S. The traffic from one
network can pass to the other, enjoying the same high-performance service.
The "(ttl=xxx!)" following the round-trip-time indicates
a different time-to-live (ttl) flag in the return packet constructed by
the router. Time-to-live is a flag at the IP packet header indicating
the "life" of the IP packet. The returned flag depends on the implementation
of routers. You can safely ignore those numbers.
Two Abilene Connected
Sites
A traceroute between the University of Illinois at Urbana-Champaign
and the University of Kansas shows the connections on an all-Abilene network.
csil-suna12% traceroute www.ukans.edu
traceroute to raven.cc.ukans.edu (129.237.33.3), 30 hops max, 40 byte packets
1 dcsgw1-csil.cs.uiuc.edu (128.174.242.1) 1.326 ms 0.958 ms
0.909 ms
2 uiuc-dcs-atm-net.cs.uiuc.edu (192.17.4.254) 1.131 ms 1.079
ms 0.995 ms
3 t-exit1.gw.uiuc.edu (128.174.1.239) 1.954 ms 1.469 ms 1.584
ms
4 dmz.gw.uiuc.edu (128.174.0.225) 2.403 ms 2.090 ms 1.926 ms
5 ipls-uiuc.abilene.ucaid.edu (192.17.10.22) 9.566 ms 8.827 ms
8.506 ms
6 kscy-ipls.abilene.ucaid.edu (198.32.8.5) 18.209 ms 17.541 ms
17.702 ms
7 ks-2-p00.r.greatplains.net (164.113.238.194) 17.754 ms 17.624
ms 17.883 ms
8 ks-2-ku.r.greatplains.net (164.113.234.205) 20.467 ms 19.710
ms 20.074 ms
9 129.237.1.254 (129.237.1.254) 20.157 ms 20.829 ms 21.404 ms
10 raven.cc.ukans.edu (129.237.33.3) 19.992 ms 20.151 ms 20.528
ms
Because both schools are on the Abilene network, the traffic
travels only on that network. The "greatplains.net" addresses indicate
the regional network used by the University of Kansas; these addresses
are on Abilene.
HPN to Commodity Site
The example below, from NCSA to the Yahoo website, shows
the difference between using a HPN and the commodity Internet.
modi4 59% traceroute www.yahoo.com
traceroute to www.yahoo.com (204.71.200.75), 30 hops max, 40 byte packets
1 lucy.ncsa.uiuc.edu (141.142.7.228) 2 ms 2 ms 3 ms
2 charlie-atm1-0-8.ncsa.uiuc.edu (141.142.11.134) 2 ms 2 ms 5
ms
3 dmz.gw.uiuc.edu (192.17.8.32) 2 ms 2 ms 2 ms
4 NChicago1-core0.nap.net (207.227.0.217) 8 ms 15 ms 6 ms
5 4.0.5.233 (4.0.5.233) 11 ms 8 ms 7 ms
6 p10-0-0.chicago1-br1.bbnplanet.net (4.0.5.89) 14 ms 6 ms 6
ms
7 h3-0-0.atlanta1-br2.bbnplanet.net (4.0.2.166) 43 ms 37 ms 49
ms
8 fa0-0-0.atlanta1-cr2.bbnplanet.net (4.0.3.250) 50 ms 43 ms
42 ms
9 s12-1-1.br1.ATL1.globalcenter.net (206.132.115.149) 114 ms
(ttl=236!) 101 ms (ttl=236!) 99 ms (ttl=236!)
10 pos2-1-155M.cr1.ATL1.globalcenter.net (206.132.115.117) 102 ms
(ttl=237!) 100 ms (ttl=237!) 104 ms (ttl=237!)
11 pos7-0-622M.cr2.SNV.globalcenter.net (206.132.151.22) 106 ms (ttl=239!)
105 ms (ttl=239!) 106 ms (ttl=239!)
12 pos1-0-2488M.hr8.SNV.globalcenter.net (206.132.254.41) 104 ms
(ttl=239!) 111 ms (ttl=239!) 102 ms (ttl=239!)
13 bas2r-ge2-0-hr8.snv.yahoo.com (208.178.103.58) 107 ms (ttl=238!) 103 ms
(ttl=238!) 110 ms (ttl=238!)
14 www10.yahoo.com (204.71.200.75) 103 ms (ttl=237!) 108 ms
(ttl=237!) 106 ms (ttl=237!)
The commodity Internet generally has a larger number of router
hops (14 machines in this example versus 6 in the first example). More
hops
means the network traffic travels a longer path through more routers, which
often results in longer delays and lower throughput. In addition, a path
traveled on the commodity Internet is nearly always provided by more than
one network service provider. In this example, two providers ("bbnplanet.net"
and "globalcenter.net") are used. Because speed, reliability, and services
depend upon all providers, getting advanced features such as Quality
of Service on the commodity Internet is nearly impossible.
2.2 ping Utility
Another useful network utility is ping, which tests
the connectivity to a remote host and reports the round-trip-time (delay)
to that host. The following example shows a ping test from a host
at NCSA to a host at SDSC.
modi4 52% ping www.sdsc.edu
PING www.sdsc.edu (198.202.75.101): 56 data bytes
64 bytes from 198.202.75.101: icmp_seq=0 ttl=250 time=57.696 ms
64 bytes from 198.202.75.101: icmp_seq=1 ttl=250 time=59.574 ms
64 bytes from 198.202.75.101: icmp_seq=2 ttl=250 time=58.229 ms
64 bytes from 198.202.75.101: icmp_seq=3 ttl=250 time=59.998 ms
64 bytes from 198.202.75.101: icmp_seq=4 ttl=250 time=61.403 ms
----www.sdsc.edu PING Statistics----
5 packets transmitted, 5 packets received, 0.0% packet loss
round-trip min/avg/max = 57.696/59.380/61.403 ms
The output (last line) shows the average round-trip-time
was 59.38 milliseconds (middle number). The minimum delay (57.696 milliseconds)
and maximum delay (61.403 milliseconds) are quite close to each other (the
difference is only 2.023), which is an indication of low delay jitter.
On a high-delay jitter path, the maximum delay could possibly be as long
as twice of the minimum delay. The delay jitter is usually considered low
when the maximum delay is no more than 10-20% longer than the minimum delay.
As with the traceroute command, you can specify
the packet size on the command line, and the exact syntax varies from platform
to platform. For example, entering ping -s 100 www.sdsc.edu on
IRIX sends out 100-byte packets. On Windows, the command ping -l 100
www.sdsc.edu does the same thing.
3. Connecting to an HPN
The computing and telecommunications unit at your campus
should be the first point of contact if you discover you are not connected
to an HPN. Staff there might be trying to document a need for a connection
so your information and interest could be of value. Campus staff are almost
always the appropriate institutional representatives to begin discussions
about connecting to an HPN.
The first step to get an Abilene connection is for your
institution to become a UCAID member. Abilene is a UCAID project. The UCAID
website has information on how
to become a UCAID member.
The High Performance Connections program at the National
Science Foundation is now winding down, but a follow-on program (or other
programs at NSF) might be available to help institutions pay initial costs
(not on-going) for physical connections. A vBNS connection grant is awarded
by NSF's Advanced Networking Infrastructure and Research (ANIR) Division,
according to the vBNS Acceptable
Use Policies (AUP). Check the vBNS
awardee list to see if any of your colleagues have been awarded a connection
grant. To apply for a vBNS grant, please read the NSF invitation.
4. Getting Help Once Connected
The National Laboratory for Applied Network Research (NLANR)
is a program funded by NSF to provide HPN support for both vBNS and Abilene
members. NLANR has three teams.
-
Distributed Application Support
Team
-
provides support for your network applications. You are welcome
to contact us for application support at dast@nlanr.net.
-
Engineering Services Team
-
provides network engineering support for your HPN connections.
Your campus network supporting staff should contact this team if they have
questions about their institutional connection, local infrastructure, or
related concerns. You can contact the engineering team for assistance at
ncne@nlanr.net or tollfree at
1-800-962-NCNE.
-
Measurement and Network Analysis
-
provides traffic measurement services. They are developing
infrastructure to monitor network conditions and publish the information
to the web. More information about measurements in provided in the Performance Section.
- DAST
Last reviewed:
December 31, 1969