Replied: Fri, 27 Oct 2000 16:10:21 -0400 Replied: erlebach@csit.fsu.edu Return-Path: erlebach@csit.fsu.edu Return-Path: Delivered-To: fox@csit.fsu.edu Received: from tarantula (tarantula.csit.fsu.edu [144.174.137.155]) by mailer.csit.fsu.edu (Postfix) with SMTP id 2486323A04 for ; Fri, 27 Oct 2000 15:59:43 -0400 (EDT) Reply-To: From: "Gordon Erlebacher" To: Subject: new_queues.doc, please comment Date: Fri, 27 Oct 2000 16:02:29 -0400 Message-ID: <00b901c04050$d5f5f000$9b89ae90@csit.fsu.edu> MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_00BA_01C0402F.4EE45000" X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook CWS, Build 9.0.2416 (9.0.2910.0) Importance: Normal X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4133.2400 This is a multi-part message in MIME format. ------=_NextPart_000_00BA_01C0402F.4EE45000 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 8bit To all users of TeraGold (the Supercomputer): Here are the current queues that you can choose from: * Interactive: 12 nodes, 30 min * Short: 40 nodes, 1 hour * Medium: 30 nodes, 24 hours * Long: 10 nodes, 10 days Note that one node has four processes. Therefore, a 12-node job may have a maximum of 48 processors. We are in the process of reevaluating these queues. We would like to give users incentive to use a large fraction of the machine, and penalize the usage of the machine in serial mode. We suggest adding a new queue named “xxx” with a maximum of 30 nodes (120 processors) and 4 days of maximum computing time. Could you please run a simple series of scaling tests for the code(s) you are running? * (fixed total problem size(): scale the number of processors * (single processor): scale the problem size This will give us an idea of the queues that are best suited to the job. Too many and too few processors will decrease the efficiency of your job and decrease overall throughput. ------=_NextPart_000_00BA_01C0402F.4EE45000 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable To all users of TeraGold (the Supercomputer):

To all users of TeraGold (the Supercomputer):

Here are the current queues that you can choose = from:

  • Interactive: 12 nodes, 30 min
  • Short: 40 nodes, 1 hour
  • Medium: 30 nodes, 24 hours
  • Long: 10 nodes, 10 days

Note that one node has four processes. Therefore, a 12-node job = may have a maximum of 48 processors.

We are in the process of reevaluating these queues. We would = like to give users incentive to use a large fraction of the machine, and = penalize the usage of the machine in serial mode. We suggest adding a new queue named = “xxx” with a maximum of 30 nodes (120 processors) and 4 days of maximum = computing time.

 

Could you please run a simple series of scaling tests for the = code(s) you are running?

  • (fixed total problem size(): scale the number of processors
  • (single processor): scale the problem size

This will give us an idea of the queues that are best suited to = the job. Too many and too few processors will decrease the efficiency of = your job and decrease overall throughput.

------=_NextPart_000_00BA_01C0402F.4EE45000--