search

codes-org / codes

The Co-Design of Exascale Storage Architectures (CODES) simulation framework builds upon the ROSS parallel discrete event simulation engine to provide high-performance simulation utilities and models for building scalable distributed systems simulations
Other
40 stars 16 forks source link

dragonfly-link-stat file: (Imported !87) #175

Closed nmcglo closed 4 years ago

nmcglo commented 5 years ago

Original MR Author: Ram Sharan Chaulagain Original MR ID: 87 Original MR URL: https://xgitlab.cels.anl.gov/codes/codes/merge_requests/87

1.router_final_function() [global connection offset was not correct] -> offset+intra-group_index 2.terminal_final_function() changed source to router and dest to terminal because LLU(s->total_msg_size) is message received by terminal

nmcglo commented 5 years ago

(2019-06-11T17:34:25.842Z): Neil McGlohon:

So this function, dragonfly_dally_terminal() is called only on a terminal LP. So any data like busy time etc is the outgoing data from terminal to router. This change will yield incorrectly labeled data.

My understanding is that there is not any output for Router to Terminal data yet as I didn't have the time to figure all that out since when I was looking at it since last week. If what you want is some Router to Terminal output data, that should go in dragonfly_dally_router_final().

nmcglo commented 5 years ago

(2019-06-11T17:36:50.670Z): Neil McGlohon:

These changes here seem good, however!

nmcglo commented 5 years ago

(2019-06-11T17:38:53.838Z): Neil McGlohon:

Hm - ok well actually i think they're both wrong (old and your new change). You're right that total msg size is amount received. But busy time is an outgoing data.

nmcglo commented 5 years ago

(2019-06-11T18:24:07.363Z): Ram Sharan Chaulagain:

Yey!!

nmcglo commented 5 years ago

(2019-06-11T18:29:31.190Z): Ram Sharan Chaulagain:

Oh yes, I just checked it and busy time is the outgoing data. And flipping variable is not the solution here it was right before but link_traffic should be changed.

nmcglo commented 5 years ago

(2019-06-11T20:10:05.012Z): Ram Sharan Chaulagain:

changed this line in version 2 of the diff

nmcglo commented 5 years ago

(2019-06-11T20:10:05.794Z): Ram Sharan Chaulagain:

added 1 commit

Compare with previous version

nmcglo commented 5 years ago

(2019-06-11T21:23:56.631Z): Neil McGlohon:

Good call here. Like the unified terminology between terminal and router state link_traffic.

nmcglo commented 5 years ago

(2019-06-11T21:26:20.857Z): Neil McGlohon:

All changes should be reflected in the reverse computation counterparts so Optimistic mode isn't broken. So you'll want to decrement the link traffic in packet_send_rc().

nmcglo commented 5 years ago

(2019-06-11T22:04:00.961Z): Ram Sharan Chaulagain:

Ya sure I will push it again with changes.

nmcglo commented 5 years ago

(2019-06-11T22:08:04.996Z): Ram Sharan Chaulagain:

added 1 commit

Compare with previous version