• Start simple: Net1Server <---->net1+gigEnn-T3+net2<----> Net2Client
  • Use one PC on each side of the wansim to avoid "confusion"
  • Avoid hubs and switches since they add it's "smartness" to the network
  • Use cross over cable to avoid "confusion"
  
  
• Setup of your wansim testing environment
  • You will need to properly calculate your Bandwidth Delay Product ( BDP = BW * rtt )
  • You probably will need to adjust your send/receive tcp buffers
    on the wansim, server, clients and hubs/switches
  • You probably will need to adjust your send/receive tcp buffers
    on your packet generators and client apps

• Flood the network for network parametric testing
  
  
• Your CPU/memory is overloaded if the wansim does NOT immediately respond to your keystrokes
• Your CPU/memory is overloaded if the wansim does NOT immediately respond html_gui changes
• Your CPU/memory/network is congested if the wansim does NOT immediately respond

• There are 5 ways to change the gigEnn WanSim parameters
  ( each way has different feature capabilities )
  • local rs232 console gui
  • remote wansim gui via ssh ( telnet ) into the wansim
  • wansim command line ( drop to shell )
  • automated shell scripts or crontabs
  • html gui ( required for MRTG and IPv6 support )
  
  
• Apply the various Network Parameters
  • gigEnn -bw 622 == 622 Mbit/s Bandwidth limit
  • gigEnn -rtt 100 == 100ms RTT delays
  • gigEnn -loss 0.001 == packet loss
  • gigEnn -qs 5183 == 5183 slot queue size
    
    
  • the various parameters can be on one line
    • gigEnn -bw 622 -rtt 100 -loss 0.0001 -qs 5183

• Is it behaving as expected
• Change your network test metrics
• Change the network and system parameters on your client/server and other hardware
• Change the network parameters on your client/server/monitor apps

• be careful NOT to overload the 800Mhz Geode CPU + 256MB system memory

Net1serverside# ping -s 30000 -i 0.01 Net2Client > /dev/null

Net1serverside# ping -s 30000 -i 0.001 Net2Client > /dev/null

• Each dot on the screen represents a dropped packet

Net1Server# iperf -w 1024K -s
Net2Client# iperf -w 1024K -c Net1Server -P 2 -t 600 == with gigEnn-t3 wansim
Net2Client# iperf -w 1024K -c Net1Server -P 10 -t 600 == with gigEnn-OC12 wansim

Net1Server# socat -u /dev/zero TCP4:Net1Client:3333,rcvbuf=64000,sndbuf=64000"
Net2Client# socat -u TCP4-LISTEN:3333,reuseaddr,rcvbuf=64000,sndbuf=64000,fork /dev/null

Net1Server# bw_tcp -s
Net2Client# bw_tcp Net1Server

Other lmbench capabilities: lat_tcp, lat_udp, tcp connection speeds, ...

Net1Server# nepim -s -w SndBuff -z RcvBuff
Net2Client# nepim -c Net1Server -w SndBuff -z RcvBuff -d
UDP: -u -W UDPSndBuff -Z UDPRcvBuff
various buffer options, ttl options, bw limits

Net1Server# netserver
Net2Client# netperf

Net1Server# netio -b 32k -s -t -u
Net2Client# netio -b 32k -t Net1Server

Net1Server# nuttcp -S -w1m

-S -6 == IPv6 server
Net2Client# nuttcp -v -w1m [options] Net1Server
-w1m == 1MB ( 1024*1024 ) window size
-u -i == UDP mode for checking packet loss
-Ri155 == 155Mbit/s limit
-Ri622 == 622Mbit/s limit

• gigEnn -rtt 5
• gigEnn -rtt 55
• gigEnn -rtt 555
  
  
• observe the RTT with ping ( ping6 ) ..
  • Net1Server# ping Net2Client

• gigEnn -bw 1540 -bwunit Kbit/s ( 1.54 Mbit/s )
• gigEnn -bw 45
• gigEnn -bw 155
• gigEnn -bw 622
  
  
• Be sure to have enough data flowing between Server and Clients
• Similarly, be sure NOT to overload the CPU/memory/network
  
  
• observe the BW limits with netstat.nn .. ( or your favorite BW monitoring app )
  • wansim# netstat.nn vr1
    
    
  • Watch out for "ran out of network buffer space"
  • Watch out for "fragmented packets"
  • Watch out for incorrect tcp buffer settings on the wansim, server, client and apps

• gigEnn -qs 0 == ( infinite queue buffers )
• gigEnn -qs 50 ( dummynet default in slots )
• gigEnn -qs 75000 -qsunit KByte ( dummynet default in Kbytes )
• gigEnn -qs 5183 -qsunit slots for 622Mbit/s * 100ms delay )
  
  

• gigEnn -loss 0 ( no packets lost )
• gigEnn -loss 0.001 ( 1 out of 1,000 packets )
• gigEnn -loss 0.000001 ( 1 out of 1,000,000 packets )
  
  
• UDP packets are easier to drop/lose
• TCP apps tries to recover its dropped/lost packets
  
  
• To force a particular amt of packet lost
  • you can probably use "these packet lost apps"
    
    
  • Use ping to send simple packets you can count, ( one packet per second )
  • Use packet lost rate of 0.1 ( 1 packet lost out of 10 ), ez enough to count n verify
  
  
• observe the packet lost with your favorite monitoring app )
  • netstat -s | grep dropped
  • ipfw pipe list look at the "dropped packet counter"

• by default, you have RTT jitter shown from ping results
• by default, you have BW jitter shown from BW monitoring tools
  
  
• To force a particular amt of jitter
  • you can probably use "these jitter apps"

• no idea how to intentionally generate collisions
  • simple collision test is to use the same IP# on different devices
  
  
• Watch the collision lights on your hubs/switches ?
• Watch for arp messages ?

• it should be trivial to overload your network and/or CPU, memory
  
  
• observe the Congestion with netstat.nn .. ( or your favorite monitoring app )
  • Kill the source of the packets overloading your network
    
    
  • if "netstat.nn" now has time to catch up to post its bandwidth data...
    you have a congestion problem
    ( the intentional flooding of the network using up the bandwidth )

• IPv6-Cloud.org IPv6 testing

root# crontab -e

  
#  
# minute hour  mday  month wday  who   command  
#  
# un-comment these to run MRTG every 5 minutes  
# --------------------------------------------  
# */5 * * * * /usr/local/bin/mrtg /home/apache/html.mrtg/wansim.cfg >/dev/null 2>&1  
# */5 * * * * /usr/local/bin/mrtg /home/apache/html.mrtg/cpu-mem.cfg >/dev/null 2>&1  
#  
#  
# simulate Bandwidth limit for T1, T3, OC-3, OC-12 at midnight, 1am, 2am, 3am ...  
#  
0 0     *     *     *     /usr/bin/gigEnn -bw 1540 -bwinit Kbit/s  
0 1     *     *     *     /usr/bin/gigEnn -bw 45 -bwunit Mbit/s 
0 2     *     *     *     /usr/bin/gigEnn -bw 155  
0 3     *     *     *     /usr/bin/gigEnn -bw 622  
0 *     *     *     *     /usr/bin/gigEnn -bw 0  
#