|
|
|
|
|
|
|
|||||||
|
|||||||||||||
|
|||||||||||||
User Group meeting, 20 July 2005
|
|
|
|||||||||||||||||||||||||||||||||||||
| > home > about > events > user_group | ||||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||||
User Group meeting, 20 July 2005 |
||||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||||
A meeting of the HPCx User Group took place at 12:30 on 20 July, 2005, using the Access Grid. Despite some technical difficulties, five sites took part, with a total of 23 people visible on screen, ten of whom were HPCx staff members.
Two presentations were given.
Using task farms on HPCx is a topic which several people have enquired about recently. David discussed the pros and cons of various techniques.
Ian discussed new features in version 9 opf xlf, which is currently installed as an NDI.
If the NDI xlf 9 causes problems for users, will we go ahead with making it the default on September 28, as planned?
Ian Bush said that this date was not completely fixed; it's important for users to try out their codes on the NDI installation before then. Any problems will be looked at carefully. Even after version 9 becomes the default, we plan to have the old version available as an NDI, although there are certain issues with regard to run-time libraries and compatibility. If we have any reason to think that going forward would cause any users problems, we will delay.
Is there any chance of a 15-minute "sanity check" queue for jobs needing 256 processors and more?
Mike Brown (Systems and Operations Team Leader) said that this presented various problems, especially for the the really large jobs (1024 processors and larger). To run such a job, it's necessary to allow a large enough space to become vacant, which can be very wasteful of time. However, it would be good to be able to do this, and we look into it; it might be possible for the smaller jobs (256 processors) at least.
If all the tasks in a task farm attempt to do IO on a single file, will this cause problems?
David Henty said that design of the GPFS file system is intended to handle this sort of challenge, and it was his understanding that there would be no problem. However, this needed to be tested.
| http://www.hpcx.ac.uk/about/events/user_group.html | contact email - www@hpcx.ac.uk | © UoE HPCX Ltd |