An easy implementation (hack) would be to wrap all of your LAM/MPI calls with a subroutine and supply two versions of the subroutine. For nodes emulating the L1 latency, the subroutine is a do-nothing routine. For those emulating the L2 latency, execute a wait loop of specified duration prior to making the LAM/MPI call. If you want, you can even tailor the latencies to be different for different classes of commands.
John Bussoletti
-----Original Message-----
From: John Korah [mailto:j_korah_at_[hidden]]
Sent: Thursday, November 13, 2003 12:01 PM
To: lam_at_[hidden]
Subject: LAM: heterogenous links simulation
Hi,
I am a beginner with LAM, so please bear with me.
I am trying to create a dual-tier network architecture
with my 64 node cluster.
This particular architecture consists of groups of
computers called supernodes. The computers within the
supernode can communicate with latency L1. the
supernode can communicate with each other with latency
L2. L2>>L1.
All the computers in my cluster are connected by a
100MBPS ethernet link.My question is that how do i
simulate different link speeds on this flat hierarchy?
I am wondering if anyone in this group has seen such implementations. If you have Suggestion/ideas, please feel free.
Thanks
John Korah
Grad Student,CS
Virginia Tech
__________________________________
Do you Yahoo!?
Protect your identity with Yahoo! Mail AddressGuard http://antispam.yahoo.com/whatsnewfree
_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
