We have been using a similar setup in "production" for our test folks since 1999 or 2000(using BSD DummyNet rather than NistNet, but the principle is the same), Being able to introduce or remove delay simply by changing IP address, not cables, is a huge win.
It is critical that when a machine is being configured to be on the "delayed" network, it is critical to correctly set ip address, subnet and default router(which would be 10.2.2.1 in Andrew's example). Even though the bits are traveling over the same physical wires, keeping the IP address spaces distinct and separate make it work with a "one-legged" router. to make management even easier, we have preloaded a handful of delay rules that apply different delay to specific small ranges of addresses on the "delayed" network, and thus users who want to pick a particular latency just do it by picking the corresponding IP address, and re-configuring their system. In this testing environment, simple fixed delay is all we need. We have actually implemented three subnets on the one NIC interface, not just two, and it all works fine. -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Andrew Moise Sent: Friday, November 17, 2006 12:38 PM To: Olle Mulmo Cc: nistnet@antd.nist.gov; Jonathon Exley Subject: Re: [nistnet] What does NISTNet as a router mean ? On 11/15/06, Olle Mulmo <[EMAIL PROTECTED]> wrote: > I believe the question was: will/can I get the expected result from having a > test setup consisting of 3 "normal" computers (i.e. 3 NICs and a separate > switch) and a bit of routing configuration magic? > > Or in any case, that's my question. :-) FWIW, we're using a setup like this (just set it up yesterday), and it seems to be working well for us. Our procedure was: * Our main network is 10.1.1.0/24. * Set up a Linux machine on that network with one NIC and two aliased interfaces (10.1.1.30 and 10.2.2.1). * Enable IP forwarding on the Linux machine. * Configure a static route in our gateway router to route traffic for 10.2.2.0/24 to 10.1.1.30. * Configure machines with a 10.2.2.0/24 address if we want them to be affected by the artifically-lagged nistnet setup. That seems to be working well for us. The killer advantage is that it doesn't take physically recabling the network in order for people to put their machines into the "lagged" environment; this is often nontrivial and would involve non-technicians poking around in patch panels and whatnot (or take up valuable IT time), which is obviously not the right thing to do. I'll let you know if we see any problems with this setup. _______________________________________________ nistnet mailing list nistnet@antd.nist.gov http://www-x.antd.nist.gov/mailman/listinfo/nistnet _______________________________________________ nistnet mailing list nistnet@antd.nist.gov http://www-x.antd.nist.gov/mailman/listinfo/nistnet
