GENIUS Meeting 7 Minutes

From RealityGrid

GENIUS project meeting 25/10/2007

Item 1:

Participants

• LSU (SJ, DK)
• UKERNA (DS)
• Leeds (JL)
• Manc (JB JM RP)
• UCL (PVC, SZ, MM, SM)
• NCSA (DM, TC)
• EPCC (SB)
• PSC (SS)

Item 2:

PVC: HPCx - 1/2 machine reserved for SC07 demos for 6 days. SB will help with scheduling.

PVC: HPCx are looking to implement advanced reservation, and make backend nodes available.

MM: HECToR - Benchmarks have been made of HemeLB on HECToR with raytracing.

PVC: We may have some time to do demos on HECToR during SC07

NGS - HARC reservations can be made on Ox and Manc, and available on Leeds too.

RP: MPI-g on NGS - done some testing this morning, still having difficulty. HemeLB and MPI-g ok, but getting Globus errors

MM: Updated HemeLB installed sites on ReG Wiki

SS: XT3 version is working and scaling as well as the XT4 version (at PSC)

PVC: with Sean we might be able to run demo on XT3 on the PSC stand at SC

JM: Bluedawg, Ducky and Zeke are production AIX machines available for SC demos

DK: Queenbee, allocated half through LONI and half through TG

SJ: will let GENIUS folks on to Queenbee when in friendly user mode. Plenty of cycles available for GENIUS users to use up

PVC: Would like to look at using the DELL clusters post SC. E.g. replica exchange which need less than 32 processors.

Item 3:

RP: There have been problems getting MPIg jobs to work on NGS. Problems occuring in the Globus layer. Got a separate installation of Globus. Hope to have it working tomorrow. Will email list to with report of problems

JL - has the stuff MF has used to build Globus, so will try to install and collaborate with MF etc

PVC - Can't test cross site MPI-g runs due to fires at SDSC.

DM: Seems like many of the problems are due to SDSC unavailability. Has sent the commands to manually cancel reservations, in response to Owain's concerns.

DM: Have made progress with problem: if job sub shortly after res made, job got stuff. Got MOAB makers involved to see if bug. Work around is to try to make reservation a bit earlier, till MOAB folk get back

PVC: When will SDSC come back?

Doru: Concern is with stability of power- best case may come back in a day or two

MM: MM and SM doing some cross site LONI runs. largely successful, some issues which should now have been resolved. SM will continue to test.

JM: TACC- emailed and asked how they want to proceed with HARC installation. Haven't heard back yet. Got email from PSC about how to install HARC on machine called Mr Rogers

JM: Wants to nag admins at NGS to apply Timezone patch and restart RMs. Not heard from Ox or Leeds - will eliminate trouble when time changes

Item 4:

PVC: UEKERNA met their own commitment to establish lightpath link. Problem with Ox terminating lightpath.

DS: Spoke to networking people at Ox and Manc. No problem with getting fibre across campus, Anthony Ryan at Manc put out a detailed proposal on how to make connection using simpler static routing. should make it quite easy. No one at Manc or Ox seems to have picked this up.

RP: Doesn't know of any progress at Manc.

DS: Project people at Manc and Ox should be pushing this.

RP: Will see what he can do at Manc.

PVC: got email from Anthony Ryan: Ox has a router, but delay with fibre. Expect to have it next Tuesday.

PVC: Will they come asking for money?

DS: Can't speak for them. Not managed to speak to anyone at Ox. Think they intend to install router then sort out cost later.

JL: Have spare lightpath going in to Leeds. Looking at how they can utilise this.

DS: Will speak to Leeds outside meeting.

JL: Could act as backup Manc -> Leeds

DS: Neil Geddis looking at making a permanent NGS lightpath backbone which will become available to all users of NGS.

Item 5:

SZ: HARC working, hope to have MPIg running early next week

JM: Zeke still has problems

SZ: Will test WS GRAM on LONI tomorrow.

SM: Have got steered MPIg app running with Viz. Will be in touch with RH about the problems. Confident it will work on Linux

SB and KR testing HemeLB etc on XT3 and XT4.

MM plans to do more benchmarks on Monday: HPCx, SDSC (if poss) & NCSA. In a few days will have more results

MM: Will try constructing bigger system to test up to 4000 cores on HECToR

MM: Need get input data from graphical editing tool to construct models from medical data. Semi automatic construction has been improved

SM: idea is to make reconstruction as automated as possible, for clinician to use.

MM: now big systems can be manipulated in a few minutes. with 8 giga-voxels and several million lattice sites (whole brain at high res)

Item 6:

SM: Part of demo will be using SPRUCE. Been in contact with Suman and others, (who've apologised for not attending). Got HemeLB compiled on Lonestar. Currently got next to run tokens, for 64 CPUs.

SJ: SPRUCE deployed on at least one machine at LSU. Should be available for use after SC

DK: Not really worked out policy issues on LONI, since don't have application to work them out with. Would like to use HemeLB as app.

Item 7:

Table on Wiki showing demos and which resources will be used.

Tim: NCSA not doing any demos in their booth.

SM: Waiting on last detail of TACC demo.

Tim: TACC guys busy getting Ranger up.

SM: Has got account on Viz workstation sorted out. would like to be on phone to someone sitting in front of workstation to test Viz on machine.

SJ: Identifies Ravi as possible person to help.

Item 8:

Next meeting next Thurs 1/11/07, then 8/11/07

PVC: Would like to make sure meetings persist after SC. Plans to continue beyond SC07

JB: Can give list of times when not available after SC.

PVC: Will arrange any fortnights. Please email preferred times.