[omnetpp] INET Ethernet tuning help

81 views

Skip to first unread message

Bradley Settlemyer

unread,

Mar 3, 2008, 2:11:23 PM3/3/08

to OMNeT++ Discussion List

Hello

I am attempting to tune my file system simulator that utilizes the INET
EtherNet package for simulating the network. My problem is that the
simulator reports a time larger to send data than my actual machines do.
For example, to send 16MB of write data to a server, my simulator requires
1.5 seconds, whereas my real machines only require 1.4 seconds.
Unfortunately, this is a fairly significant difference, as 1.5 seconds falls
well outside of my confidence interval using a 1 sample t test against the
simulator derived mean. Can anyone please, please, please suggest a way to
make the simulator go faster (It is slower for write sizes of 1MB, 512K, and
64K as well).

I have tried cranking up many of the EtherSwitch settings, but that has
had very little impact. Here are the settings I have that I think are
relevant:

# We should set this to something like a Foundry FastIron FESX448
adenine.switch.relayUnitType = "MACRelayUnitPP"
adenine.switch.relayUnit.addressTableFile = ""
adenine.switch.relayUnit.addressTableSize = 100000
adenine.switch.relayUnit.agingTime = 0
adenine.switch.relayUnit.processingTime = 0
adenine.switch.relayUnit.bufferSize = 262144
adenine.switch.relayUnit.highWatermark = 100000
adenine.switch.relayUnit.pauseUnits = 0
adenine.switch.relayUnit.writeScalars = true
adenine.switch.mac[*].address = "auto"
adenine.switch.mac[*].txQueueLimit = 128000
adenine.switch.mac[*].txrate = 136000000000
adenine.switch.mac[*].duplexEnabled = true
adenine.switch.mac[*].writeScalars = true

# We should set this to something like an Intel Pro/100 card
#
# RX Ring Length: 256 TX Ring Length: 128
#
adenine.**.eth[*].queueType = "DropTailQueue"
adenine.**.eth[*].queue.frameCapacity = 1000
adenine.**.eth[*].mac.promiscuous = false
adenine.**.eth[*].mac.address = "auto"
adenine.**.eth[*].mac.txrate = 100000000
adenine.**.eth[*].mac.duplexEnabled = true
adenine.**.eth[*].mac.txQueueLimit = 1000
adenine.**.eth[*].mac.queueModule = ""
adenine.**.eth[*].mac.writeScalars = true
adenine.**.eth[*].encap.writeScalars = true
adenine.**.ppp[*].queueType = "DropTailQueue"
adenine.**.ppp[*].queue.frameCapacity = 0

My goal is to publish some results in April, but I'm afraid without some
sort of statistical validation of my model, I will have a difficult time
getting accepted to a quality conference.

Cheers,
Brad
_______________________________________________
OMNeT++ Mailing List
options: http://lists.omnetpp.org/mailman/listinfo/omnetpp-l
archive: http://www.omnetpp.org/listarchive/index.php

Bradley Settlemyer

unread,

Mar 3, 2008, 9:45:01 PM3/3/08

to OMNeT++ Discussion List

I switched to the EtherSwitch2 (and EtherInterface2) model with manually
configured 100Mbit channel, and that took about 40 milliseconds of delay
off. I'm still about 80 milliseconds slower in simulation than in reality.
Any advice would be extremely welcome.