Remote/Monitored host
There are two possibilities. One requires the administrator of the
remote host to install required servers, the other requires the
administrator to set up an account (iepm) and password for us so
we can install the remote host toolkit.
Admin at Remote site installs servers
The host requirements are roughly:
CPU: >= 1000MHz, NIC card >= 100Mbits/s full duplex (GE preferred),
64 bit 66MHz PCI bus. If site has good connectivity to the Internet
(e.g. >= OC3), then a GE interface is necessary.
The host SHOULD be registered with DNS.
- The host must be configured for large TCP buffers/windows, see
TCP Tuning Guide.
- Install servers for:
iperf (version 2.0.2 or later),
thrulay,
pathchirp and
pathload.
- Start the servers, and ensure they keep running (e.g. have a cron job test
they are running and if not restart, also make sure start after re-boot).
- Open TCP ports 5000 (iperf),5003 (thrulay), 55002 (pathload) and UDP
ports 8365 (pathchirp) and 55001 (pathload).
- Unblock pings and enable traceroute to the remote host.
- Install a
reverse traceroute server, following the instructions
here.
- Contact us
with the host name,
the ports that have been opened, details on the remote host (name, hardware
configuration, NIC speed, OS), the site's location (city and/or
latitude/longitude),
connectivity to the Internet and the URL for the traceroute server.
SLAC installs remote host toolkit
We need an account ideally with the with the name of iepm, running
under the tc shell (tcsh),
on a suitable Linux or Solaris
host at your site. The host must also support ssh.
The $HOME directory must not be
in AFS space, NFS or local space is fine.
CPU: >=
1000MHz, NIC card >= 100Mbits/s full duplex (GE preferred),
64 bit 66MHz PCI bus. If site has good
connectivity to the Internet (e.g. >= OC3), then a GE interface is necessary.
Disk space to save a file copy of up to 2GBytes.
Ideally this disk space should be mounted so the file cannot be cached and it
must not be /tmp space mapped to swap/memory space.
Operating system: Linux 2.4 or later
Iperf processes appear to take (according to top) about 5MB, there should
only be one iperf process running at a time on the remote host for IEPM.
In addition they need roughly 2*streams +1 window buffers.
Typical window buffers * streams are about 4MBytes for each process.
System tools: ssh protocol version 1 or 2, gunzip, tar, make
Tools: pico, C++ (c++ or g++), C (gcc or cc), Perl 5, Pthreads library, 64
bit libc support, libz
Need an account with userid cottrell with ssh access & RSAAuthentication
yes in /etc/ssh/ssh_config, does not need root.
TCP ports 22 (for ssh), 5000-5011 (for iperf, 5000 is used most frequently,
5001-5011 are used for occasional measurenents with varying window sizes,
each port in the range 5001-5011 uses a different maximum receive window size),
5003 for
thrulay,
5020-5022 (for bbftp, most of the traffic will be seen on 5020, 5021 is the client's
default control port and 5022 is the server control port),
5031-5039 (for bbcp, it tries 5031 first and only if it is already in use
will it try a different port in the specified range),
and 2811 (for GridFTP) must be open (inbound to the site) from the outside world.
Also UDP port 8365 for pathchirp,
UDP 55001 and TCP 55002 for pathload and port 8176 must be opened for ABwE/abing.
We also need to know the speed of the NIC, and the speed of the site's
connection to the Internet. If there are any unusual features such as the host
is multihomed (in which case we need to know the IP address of the
interface to use), it may be important to let us know.
We will create the following directories:
$HOME/bin, $HOME/package, $HOME/pinger/load. We will also set up a .forward
file to forward email to the account.
Click here to find instructions for
conducting a localmake on remote hosts.
Network Impact
A set of measurements are made to each remote host about every 90 minutes.
Each set typically consists of a set of 10 pings, a traceroute,
an ssh (port 22) to start the iperf client, a 10 second
iperf TCP client (usually port 5000) run to measure throughput, a 10 second
bbcp memory-to-memory run, a 10 second bbcp disk-to-disk run, and a 10
second bbftp run. The traffic generated by the
iperf client, bbcp and bbftp runs depends on the bottleneck achievable
throughput of the link. See
SLAC WAN bandwidth Measurement Tests for some examples.
Monitoring Host
The monitoring host has the same requirements as the remote host, plus:
- Monitoring hosts must have access to the perl libraries listed here.
- There must be a version of gnuplot available.
- The analysis scripts must be able to write the web reports to a
file system that is accessible to a web server.
- Cron jobs must be able to send email to the user account on the host.
- All monitoring hosts MUST be registered with DNS.
- The host must be configured for large TCP buffers/windows, see
TCP Tuning Guide.
- Disk space:
- At SLAC we are using about 500MBytes for all the files accessible to the
web server (plots, tables, narrative, extract, and analyzed
data files).
- We need about 200MBytes for files to transfer/copy by the
sensors. This should be local
to the monitoring host so we are not measuring the speed of the local network.
- We need about 150MBytes for scripts, configuration files, keys and
executables. This should NOT be readable by the web server.
- We need up to ~4MBytes/month/remote host to save the log data.
This should NOT be readable by the web server.
- If either or both of the executable or log file or web
spaces are in AFS then we need a way to renew the
token for the cron jobs.
Demo Host
The demo host for iGrid2002 and SC2002 needs all the above plus:
Created November 2, 2001, last update July 11, 2002.
Comments to iepm-l@slac.stanford.edu
