[Gems-users] Stats on bandwidth overhead?

[Calvin <calvincv@xxxxxxxxx>]

Hello,

I am running a simulation with the MSI_MOSI_CMP_Directory protocol and need
to measure the bandwidth overhead after changing some message size.
Could you help clarify how to check the results?

Specifically, I saw the following section in the Ruby stats file. But
I don't quite
understand this. For example, how could the link utilization for
switch_0_link_0
be larger than 1? For evaluating the bandwidth overhead, should I use
the average
link utilization, or the ratio of total number of bytes transferred by
the switch to
the execution time?

Network Stats
-------------

switch_0_inlinks: 5
switch_0_outlinks: 5
links_utilized_percent_switch_0: 0.695003
 links_utilized_percent_switch_0_link_0: 1.02101 bw: 100000 base_latency: 1
 links_utilized_percent_switch_0_link_1: 0.241305 bw: 100000 base_latency: 1
 links_utilized_percent_switch_0_link_2: 1.49691 bw: 100000 base_latency: 1
 links_utilized_percent_switch_0_link_3: 0.715788 bw: 10000 base_latency: 40
 links_utilized_percent_switch_0_link_4: 0 bw: 100000 base_latency: 1

 outgoing_messages_switch_0_link_0_Control: 10097 80776 [ 0 0 1066
9031 0 ] base_latency: 1
 outgoing_messages_switch_0_link_0_Data: 12559 904248 [ 0 0 0 12559 0
] base_latency: 1
 outgoing_messages_switch_0_link_1_Control: 1398 11184 [ 0 0 957 441
0 ] base_latency: 1
 outgoing_messages_switch_0_link_1_Data: 3078 221616 [ 0 0 0 3078 0 ]
base_latency: 1
 outgoing_messages_switch_0_link_2_Control: 17673 141384 [ 15639 0
1096 938 0 ] base_latency: 1
 outgoing_messages_switch_0_link_2_Data: 18094 1302768 [ 9472 0 0
8622 0 ] base_latency: 1
 outgoing_messages_switch_0_link_3_Control: 8632 69056 [ 0 8632 0 0 0
] base_latency: 40

switch_1_inlinks: 1
switch_1_outlinks: 1
links_utilized_percent_switch_1: 0
 links_utilized_percent_switch_1_link_0: 0 bw: 100000 base_latency: 1
-------------------------------------------------------------------------------------------------------------------

Thanks,
Calvin