[codes-ross-users] Fat-tree configuration support oversubscription >1?

[noah wolfe]

Hi Peixin,

The current implementation of the CODES fat-tree model only supports 1:1 
port mapping for leaf level switches to maintain the standard full 
bisection fat-tree layout (i.e. half the ports go down to hosts and the 
other half go up to the next level switches). However, the mapping of 
ports can be adjusted for leaf level switches (and all levels of 
switches) in the switch_init() function. Perhaps adding another ft_type 
for oversubscription of leaf switches would be the best approach. The 
mapping of hosts to switches which is done in the ft_terminal_init() 
would also need to be adjusted for the increased number of host 
connections to leaf switches. Both functions are found in 
src/networks/model-net/fattree.c.

By the way, recently updated documentation for the CODES fat-tree model 
can be found on the wiki: 
https://xgitlab.cels.anl.gov/codes/codes/wikis/codes-fattree

Thanks,
Noah


From: <codes-ross-users-bounces at lists.mcs.anl.gov 
<mailto:codes-ross-users-bounces at lists.mcs.anl.gov>> on behalf of Peixin 
Qiao <pqiao at hawk.iit.edu <mailto:pqiao at hawk.iit.edu>>
Date: Tuesday, January 31, 2017 at 12:02 PM
To: "codes-ross-users at lists.mcs.anl.gov 
<mailto:codes-ross-users at lists.mcs.anl.gov>" 
<codes-ross-users at lists.mcs.anl.gov 
<mailto:codes-ross-users at lists.mcs.anl.gov>>
Subject: [codes-ross-users] Fat-tree configuration support 
oversubscription >1?

Hi there,

When I do research about fat-tree topology uisng CODES, the 
fat-tree.conf shows that the topology is with an oversubscription of 
1:1, which indicates that all hosts may potentially communicate with 
arbitrary other hosts at the full bandwidth of their network interface. 
But the Stampede systems, which I simulated is with an oversubscription 
of 1.25:1, which means that each leaf switch with 36 ports is directly 
connected to 20 hosts and 16 ports in the core switch. So, does CODES 
support this configuration? And how should I modify the configuration file?

Best Regards,
Peixin Qiao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/codes-ross-users/attachments/20170201/1552032a/attachment.html>