READMEpathload

$Header: /net/cvs/bwtest/pathload/README,v 1.32 2006/05/19 22:34:59 jain Exp $

Pathload Release  : Ver 1.3.2

Contents
-------- 
Overview of Pathload
Building Pathload
Running Pathload
Examples
Interpreting Pathload output
Contact and other info


Overview of Pathload
--------------------
Pathload is a tool for estimating the available bandwidth
of an end-to-end path from a host S (sender) to a host R (receiver). 
The available bandwidth is the maximum IP-layer 
throughput that a flow can get in the path from S to R, 
without reducing the rate of the rest of the traffic in the path. 

The estimation algorithm of Pathload is based on the SLoPS methodology,
which is described in: 
"Pathload : A measurement tool for end-to-end available bandwidth",
by Manish Jain and Constantinos Dovrolis, published at PAM 2002.
That paper is available at:
http://www.cc.gatech.edu/~dovrolis/Papers/pam02.ps
A more extensive study of available bandwidth is given in:
"End-to-End Available Bandwidth: Measurement methodology, Dynamics, and 
Relation with TCP Throughput", in Proceedings of ACM SIGCOMM in August 2002. 
That paper is available at:
http://www.cc.gatech.edu/~dovrolis/Papers/sigcomm02.ps.gz

Briefly, Pathload is based on the following key idea: 
When a process at S sends a periodic stream of UDP packet 
at a rate higher than the available bandwidth in the path,
the relative one-way packet delays show an increasing trend.
When the stream rate is lower than the available bandwidth, the 
relative one-way packet delays show no consistent trend.

Pathload consists of a process running at S and a process running at R.
S sends periodic streams of UDP packets from S to R at a certain rate. 
Pathload does not determine whether a particular rate (Tr) is 
greater than available bandwidth (A) based on just one stream.
Instead, it sends "a fleet of N streams", so that it has N samples
to decide whether Tr > A, or not.

Upon the receipt of a complete fleet, R checks if there is an increasing 
trend in the relative one-way packet delays in each stream. 
If a large fraction f of the N streams in a fleet show increasing trend,
then the entire fleet is said to have increasing trend and the
next fleet rate is lower than Tr.
If a large fraction f of streams in a fleet show no trend, then the 
entire fleet is said to have no trend and the next fleet rate is higher 
than Tr.

A certain fleet rate falls into the `grey region' when less than f x N 
streams of the fleet show increasing trend, and less than f x N streams 
show no trend.
This can happen when the available bandwidth was more than fleet rate
for some streams and less than the fleet rate for some other streams.
In other words, the available bandwidth during the fleet duration varied
above and below the fleet rate, causing some streams to show increasing 
trend and others to show no trend. 

The iterative estimation algorithm of Pathload terminates when:
- the rate of two successive fleet is less than a user-specified resolution,
- or, the available bandwidth varies in a `grey-region', which is larger
than the user-specified resolution. 


Building Pathload
-----------------
Pathload uses the standard configure/make approach.
$> ./configure
$> make

After you have extracted the Pathload code in a directory,
run configure in that directory. Then run make.

You should then have two executables in the Pathload directory.
1. pathload_snd (to be run at the sender)
2. pathload_rcv (to be run at the receiver)



Running Pathload
----------------

sender 
   To do available bandwidth estimation, first run pathload_snd at the sender S:
   $> pathload_snd [-i] [-h|-H] [-q]

   [-i] By default, pathload_snd will exit after 1 measurement.
   Use switch -i to run sender in iterative (persistent) mode.
   
   [-q] By default, sender sends minimal output (fleet rate) to terminal 
   (stdout). To completely disable any message on stdout,
   use -q (quite mode).  

receiver 
   Then, run pathload_rcv at the receiver R:
   $> pathload_rcv [-q|-v] [-o|-O <filename>] [-N <filename>][-w <bw_resol>] 
      [-h|-H] [-t <timeout in sec>] -s <sender>
        -s        : hostname/ipaddress of sender
        -q        : quite mode
        -v        : verbose mode
        -w        : user specified bandwidth resolution
        -o <file> : write log to user specified file [default is pathload.log]
        -O <file> : append log to user specified file [default is pathload.log]
        -N <file> : append output in netlogger format to <file>
        -t <time> : timeout value in seconds
        -h|-H     : print this help and exit
      
   [-o] By default, receiver always appends detailed output to file "pathload.log". 
   The output file can be changed by specifying a filename with -o switch. Note that 
   if the specified file already exits, it will be over-written.

   [-O] Similar to -o except that the log is appended to the file.
   
   [-q,-v] By default, receiver sends minimal output (fleet rate and final 
   result) to terminal (stdout). To completely disable any message on stdout,
   use -q (quite mode).  To enable detailed output on screen, use -V (Verbose
   mode). Note that, log file (user specified or "pathload.log") contains 
   exactly the same output as -V.
   
   [-N] Specify a file name to store output in netlogger compatible format.

   [-w] By default, bandwidth resolution (bw-resol) is set to a fraction
   of the Average Dispersion Rate (ADR) in the path. 
   To change default bw-resol use -w.  bw-resol allows the user to select 
   the precision with which Pathload should try to estimate the available 
   bandwidth. Keep in mind that the available bandwidth is a dynamically varying
   metric, and so its variation range may be wider than the user-specified 
   resolution. In that case, Pathload will report the "grey region",
   which is an estimate of the variation range of the available bandwidth. 

   [-t] Specified the timeout value. If the total execution time of the process 
    exceeds the timeout value, terminate the pathload receiver.

EXAMPLES
--------
Say you want to measure the available bandwidth from 
host A to host B. 
Run pathload_snd at A and pathload_rcv at B.

At host A,
$> ./pathload_snd

At host B,
$> ./pathload_rcv -h A 
In this example, the sender will terminate after the receiver completes one
measurement. 
The detailed output will be logged to default file "pathload.log".

Now, if you wanted to change the default log file at receiver B:
$> ./pathload_rcv -h A -o testrun.log

If you also wanted detailed output on screen as well netlogger output:
$> ./pathload_rcv -h A -o testrun.log -N netlogger.log -V



Interpreting the Pathload output
--------------------------------

Sample default output on receiver :-
Line # | Actual output
~~~~~~~~~~~~~~~~~~~~~~
Line 1 : jawks [507]$ ./pathload_rcv -h sirius
Line 2 : Receiver jawks starts measurements on sender sirius at Sun Apr \
         6 15:29:44 2003
Line 3 : Receiving Fleet 0, Rate 97.40Mbps
Line 4 : Receiving Fleet 1, Rate 193.55Mbps
Line 5 : Receiving Fleet 2, Rate 145.47Mbps
Line 6 : Receiving Fleet 3, Rate 121.44Mbps
Line 7 : Receiving Fleet 4, Rate 109.42Mbps
Line 8 : Receiving Fleet 5, Rate 103.41Mbps
Line 10:         *****  RESULT *****
Line 11: Available bandwidth range : 97.40 - 103.41 (Mbps)
Line 12: Measurements finished at Sun Apr  6 15:29:44 2003
Line 13: Measurement latency is 5.69 sec

Explanation :-
     Line 1 Shows the command executed for this run of Pathload.
     Line 2 Shows the sender and receiver hostname and start time of 
            measurements.
     Line 3-8 Show the various rates that Pathload tries iteratively
            to estimate avail-bw range.
     Line 11 Shows the final estimate of available bandwidth range.
     Line 12 Shows the finish time.
     Line 13 Gives the total time taken for measurement. 


Sample detailed output on receiver (same as in log file):
(not a complete snapshot, but shows all the different message output)
  
Line 1 : Receiver jawks starts measurements on sender sirius at Sun Apr \
         6 15:29:44 2003 
Line 2 :   Maximum packet size          :: 1472 bytes
Line 3 :   send latency @sndr           :: 4 usec
Line 4 :   recv latency @rcvr           :: 5 usec
Line 5 :   Minimum packet spacing       :: 10 usec
Line 6 :   Max rate(max_pktsz/min_time) :: 1200.00Mbps
Line 7 :   ADR [.]                      :: 97.40Mbps
Line 8 :   Grey bandwidth resolution    :: 4.87
Line 9 : 
Line 10: 
Line 11: Receiving Fleet 0
Line 12:   Fleet Parameter(req)  :: R=97.40Mbps, L=946B, K=100packets, T=80usec
Line 13:   Lossrate per stream   :: :0.0:0.0:0.0:0.0:0.0:0.0:0.0:0.0:0.0:0.0:0.0:0.0
Line 14:   # of CS @ sndr        :: : 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0
Line 15:   # of CS @ rcvr        :: : 1: 0: 0: 1: 0: 0: 0: 0: 2: 0: 1
Line 16:   # of DS @ rcvr        :: : 1: 0: 0: 1: 0: 0: 0: 0: 2: 0: 1
Line 17:   Fleet Parameter(act)  :: R=93.88Mbps, L=946B, K=100packets, T=83usec
Line 18:   PCT metric/stream[12] :: 0.09:0.27:0.27:0.36:0.09:0.09:0.18:0.27:0.36:0.09:0.09:0.20:
Line 19:   PDT metric/stream[12] :: 0.33:0.28:0.00:0.68:-0.14:-0.11:-0.20:-0.03:0.41:-0.33:-0.50:0.12:
Line 20:   PCT Trend/stream [12] :: NNNNNNNNNNNN
Line 21:   PDT Trend/stream [12] :: NNNINNNNUNNN
Line 22:   Trend per stream [12] :: NNNUNNNNNNNN
Line 23:   Aggregate trend       :: NO TREND
Line 24:   Rmin-Rmax             :: 97.40-0.00Mbps
Line 25:   Gmin-Gmax             :: 0.00-0.00Mbps
Line 26: 
Line 27: 	*****  RESULT *****
Line 28: Available bandwidth range : 99.59 - 102.91 (Mbps)
Line 29: Measurements finished at Sun Apr  6 15:37:02 2003 
Line 30: Measurement latency is 5.83 sec 

Explanation :-
     Line 2 Minimum of of the maximum packet sizes on sender and receiver.
     Line 3 Latency of "send" system call at sender.
     Line 4 Latency of "recv" system call at receiver.
     Line 5 Two times the maximum of "send" and "recv" latency.
     Line 6 Maximum rate which can be measured.
     Line 7 Asymptotic dispersion rate, used as the rate of first fleet.
     Line 8 Grey bandwidth resolution.
          
     Line 11-25 Information corresponding to each fleet 
     Line 11 Shows the fleet number.
     Line 12 Shows the fleet parameters that the receiver requested from the
             sender: 
              R (rate of fleet)
              L (packet size)
              K (number of packet per stream)
              T (inter-packet spacing)
     Line 13 Shows the loss rate per stream, with ":" used as separator.
     Line 14 Shows the number of instances when sender side processes was 
             "possibly" interrupted while sending stream.
     Line 15 Shows the number of instances when receiver side processes was 
             "possibly" interrupted while receiving stream.
     Line 16 Shows the number of packets received when receiver side process
             was interrupted.
     Line 17 Shows the actual fleet parameters (same format as in line 2).
     Line 18 Shows the actual PCT metric for relative one-way packet delay 
             trend per stream.
     Line 19 Shows the actual PDT metric for relative one-way packet delay
             trend per stream.
     Line 20 Shows the one-way delay trend, per stream, using PCT.
              I  - Increasing trend.
              N  - No trend.
              U  - Unusable.
     Line 21 Shows the one-way delay trend, per stream, using PDT.
              I  - Increasing trend.
              N  - No trend.
              U  - Unusable.
     Line 22 Shows trend of stream, after combining result of PCT and PDT.
              I  - Increasing trend.
              N  - No trend.
              U  - Unusable.
     Line 23 Shows the aggregate trend of the fleet. Possible values are:
              INCREASING 
              NOTREND
              GREY
     Line 24 Shows current value of (Rmin, Rmax) after current fleet.
              Rmin - maximum rate which showed No trend.
              Rmax - minimum rate which showed Increasing trend.
     Line 25 Shows current value of (Gmin, Gmax) after current fleet.
              Gmax - maximum rate (< Rmax) which showed grey region.
              Gmin - minimum rate (> Rmin) which showed grey region.
     Line 27-30 Result of Pathload
     Line 28 Available bandwidth range during the measurement period
     Line 30 Total time taken for measurement

Finally, Pathload exits with one of the following messages :-

1. Avail-bw < minimum sending rate.
   Increase MAX_TIME_INTERVAL in pathload_rcv.h from 200000 usec to a
   higher value.

   Sending rate of stream is determined by packet size and spacing
   between packets.  MAX_TIME_INTERVAL in pathload_rcv.h is currently
   set to 200000 usec.  If you want to probe at rates lower than
   possible with current version of tool, increase this value to
   higher and recompile the code.  

2. Avail-bw > maximum sending rate.

   This constraint limits the maximum rate that Pathload can generate.
   Maximum rate depends on the maximum packet size and time latency of
   send/recv call.  If that rate is reached at a certain fleet, the
   tool does not proceed any further.
   
3. Measurement terminated due high rate of CS @ sender/receiver.
   
   It implies that either the sender or receiver host was interrupted
   while sending a single stream. This could be due to a lot of CPU
   intensive processes running on either host.  In the log file, two
   fields "# of CS @ sndr" and "# of CS @ rcvr" will give some
   indication as to which host is causing context switching in
   Pathload.

   Another reason may be, if you are sure that your machine is lightly
   loaded, that packet stream experienced interrupt coalescing (IC) at
   the receiver side NIC.  The current algorithm in pathload includes
   algorithm to detect IC, however it has been tested on NICs from two
   vendors. Since vendors may implement   IC differently, IC may be
   incorrectly interpreted as CS.

4. Exiting due to user specified resolution.

   The user specified resolution could be achieved.

5. Exiting due to grey bw resolution.

   Gmin and gmax range determine the grey bw resolution and pathload
   determines when the range is close enough to report as final
   result. The range of gmin-gmax gives an indication of variation
   range of available bandwidth during measurement period.

6. Actual probing rate != desired probing rate.

   When Pathload exits with this message, it means that during at
   least one fleet, the sender was unable to send a fleet with  the
   inter-packet spacing that the receiver requested. 

7. Receiver NIC has interrupt coalescence enabled.

   The message implies that NIC at receiver does interrupt coalescence
   (IC). In brief, effect of interrupt coalescence is to destroy the
   one way delay in packet stream. pathload tries to detect IC and
   reports the lower bound on avail-bw. Details about effects of IC on
   pathload are described in our PAM 2004 paper "Effect of Interrupt
   Coalescence on Network Measurements".

Authors
-------
1. Manish Jain
   jain@cc.gatech.edu
2. Constantinos Dovrolis
   dovrolis@cc.gatech.edu


Acknowledgments
---------------
This work was supported by the SciDAC program of the US Department 
of Energy (award # DE-FC02-01ER25467).