titan

Up since 11/8/17 02:45 pm

eos

Up since 11/14/17 11:20 pm

rhea

Up since 10/17/17 05:40 pm

hpss

Up since 11/20/17 09:15 am

atlas1

Up since 11/15/17 07:25 am

atlas2

Up since 11/27/17 10:45 am
OLCF User Assistance Center

Can't find the information you need below? Need advice from a real person? We're here to help.

OLCF support consultants are available to respond to your emails and phone calls from 9:00 a.m. to 5:00 p.m. EST, Monday through Friday, exclusive of holidays. Emails received outside of regular support hours will be addressed the next business day.

Loading Events

OLCF User Call: Thread Affinity on Rhea and Eos

Details

Date:
June 3, 2014
Time:
12:30 pm - 1:30 pm

The OLCF will be resuming the monthly User Conference Call. These calls, which will be held at 12:30 PM Eastern Time on the first Tuesday of each month, are your opportunity to speak with center personnel to get the latest updates, express any concerns you may have, etc.

Monthly Topic: Thread Affinity on Rhea and Eos
We will give an overview of thread affinity on Eos and Rhea and discuss common pitfalls that affect thread layout.
Speaker: Robert French of OLCF/UAO
Slides: Thread Affinity,OLCF-News6-3-14

Streaming recording link:
https://olcf.webex.com/olcf/ldr.php?RCID=080cb7369484772e6d59f30c3744a3d2
Download recording link:
https://olcf.webex.com/olcf/lsr.php?RCID=fa7cf77b6a4572641716624a6fc7282a

Summary
We had several good questions.
1. One was about core specialization (-r 1) and also using the -j flag to control thread affinity.
A related talk which covers this for Eos and Titan can be found here:aprun usage.

2. Q. What about helper threads? Can they be controlled?
A. Helper threads, are spawned to assist main thread computation by prefetching memory from various shared caches. They do not execute computation. In some cases they can compete with the main thread for cache space and bandwidth. Generally they can not be controlled by the programmer.

3.Q. Can the -cc numa_node be the default at runtime?
A. -cc numa_node assigns ranks to a particular numa node. We can not make this the default because it would not work for all codes.
For more information on Numa nodes see:
Titan Task Layout
Eos Task Layout

4. Q. Is there a better method to detect the thread architecture automatically at run time?
A. We don’t have one at this time.

If you want to download a recoding of the whole call please click “Full Recording“.

If you are at ORNL You may attend in person in Bldg. 5600 C101.

To attend by phone
call 1-855-244-8681
Access code:627 104 628

To attend by WebEx:
https://olcf.webex.com/olcf/j.php?MTID=m73a0528f65437aa1c9d05e2e6f725a09
Enter the meeting password: users
Click “Join”.
For usage instructions, see
https://www.olcf.ornl.gov/kb_articles/using-cisco-webex-at-olcf/