Minutes from 12/03/2013

December third, 2013 at 10am in Room N104 Coverdell

Agenda for 12/03/2013

Members Present: Jonathan Amster, Jonathan Arnold, Martha Carr, Peter Kner, Jim Leebens-Mack, and David Saltz

Members Absent: Margie Lee, Richard Watson, and Yiping Zhao

Ex-Officio Members Present: Jessie Kissinger, Thiab Taha, Nick Berente, Michael Lucas, Bob Scott, and Guy Cormier

Guests: Ping Ma

• The minutes of the October 29th meeting were unanimously approved.
• The rest of the meeting was taken up with a presentation by Guy Cormier on the upcoming ~$1.5M purchase. Supporting documents are linked below.
  • There was a discussion of high memory nodes. The data assembled by Guy shows that most (~70%) jobs on high memory (512Gb) nodes do not use more than 16Gb of memory. The new general purpose nodes will have 128Gb. There are some jobs that use over 1Tb of memory (when virtual memory is included). The purchase of private nodes and cost sharing were both discussed. The Advisory Committee agreed that there should be at least one 1Tb high memory node in the order although no formal vote was taken.
  • Next there was a discussion of GPU nodes. The current GPU nodes are used by 5 or 6 research groups at maximum capacity. The Advisory Committee unanimously voted to include one GPU Fat Node (8x Nvidia Kepler K40) in the purchase. It was felt that this would help with the upcoming NSF MRI proposal led by Ping Ma.
  • Currently there is 60Tb of usable scratch memory. This is part of the Panasas system and will be used for Home directories (Tier 2). Guy recommended the purchase of 96 Tb (80 usable) of new scratch memory. It was felt that this was not a large enough increase to account for growth although the scratch space is currently not used in an optimal manner. Ideally, after jobs are done data should be removed from scratch to home directories. Jessie brought up the fact that this can be slow, especially if it needs to be transferred back and forth multiple times. There was a brief discussion of purchasing the highest end equipment (solid state drives, etc.) to increase transfer speeds. The Advisory Committee unanimously voted to purchase 3 48Tb chains for scratch space.
  • Currently, approximately 200Tb of archive storage (Tier 3) are being used. This has not been widely advertised so perhaps there would be a lot more use if it was better known. Jessie felt that at least 1Pb is needed and a few people said they thought that having a Petabyte of storage would help with the NSF MRI proposal. Ping Ma would like to make "Big Data" part of the proposal. The current ~320Tb of archival storage is not under warranty, and Guy would like to replace it to increase the security of archival storage. By purchasing 3 320Tb chains of archival storage, the GACRC would have nominally 1Pb of storage. The Advisory Committee unanimously voted to purchase 3 320Tb chains for archival space.
  • Bob Scott made the point that the total cost of the purchase should be discussed to make sure that the GACRC has enough funds for operating over the next three years.

File:GACRC AC Dec3.pdf, File:NewClusterWorksheet.pdf