Parallel Workloads Archive: SDSC DataStar

The San Diego Supercomputer Center (SDSC) DataStar log

System:	184-node IBM eServer pSeries 655/690
Duration:	Mar 2004 thru Mar 2005
Jobs:	96,089

A log covering about a year of activity. It contains information on the requested and used nodes and time, CPU time, submit, wait and run times, and user.

The original log, with all the available information, is (was) available directly from the NPACI JOBLOG repository, in a special format described on that site. The copy available here is in the Standard Workload Format, which loses some data as specified below.

The NPACI JOBLOG repository is made available by Victor Hazlewood. If you use this log in your work, please use a similar acknowledgment.

This log is subject to the NPACI JOBLOG Repository Data Usage Agreement:
This Job Trace Repository is brought to you by the Allocated Systems and Security Technologies group of the San Diego Supercomputer Center (SDSC).
The JOBLOG and job trace data is Copyright 2000-2005 The Regents of the University of California All Rights Reserved
Permission to use, copy, modify and distribute any part of the JOBLOG data or job trace information from systems at the San Diego Supercomputer Center for educational, research and non-profit purposes, without fee, and without a written agreement is hereby granted, provided that this copyright notice is preserved in all copies or directories containing the data and all works based on use or analysis of this data is properly referenced in any written or electronic publication.

Downloads:

SDSC-DS-2004-2.swf	1.6 MB gz	converted log
SDSC-DS-2004-2.1-cln.swf	1.6 MB gz	cleaned log -- RECOMMENDED, see usage notes
SDSC-DS-2004-1.swf	1.6 MB gz	OLD VERSION of converted log (replaced 6 Dec 2011)

(May need to click with right mouse button to save to disk)

Papers Using this Log:

This log was used in the following papers: [sotomayor06] [feitelson07a] [feitelson08] [folling09] [aida09] [liuz10] [lindsay12] [krakov12] [zakay12] [zakay13] [chen13] [rajbhandary13] [sheikhalishahi14] [zakay14] [zakay14b] [feitelson14] [lic14] [ntakpe17] [hai20]

System Environment

The total machine size is 184 nodes.

The nodes are of either of two types: p655, an 8-way SMP with 1.5 GHz processors and 16GB of memory, or p690, a 32-way SMP with 1.7 GHz processors and 128 GB of memory. These nodes are divided into 3 partitions as follows:

# Partition p655 p690

1 interactive 5 1

2 batch 171 0

3 batch 0 7

(the p690 node in the interactive partition has only 64 GB of memory.)
Jobs are submitted to a set of queues. The main ones are

Name Nodes Time limit Node limit When

interactive p655

interactive32 p690

express p655 2hr 8nodes weekdays

high p655 18hr 164nodes

normal p655 18hr 164nodes

express32 p690 2hr 7nodes weekdays

high32 p690 18hr 7nodes

normal32 p690 18hr 7nodes

expressL p690 2hr 1nodes weekdays

highL p690 18hr 1nodes

normalL p690 18hr 1nodes

TGexpress p690 2hr 7nodes weekdays

TGhigh p690 18hr 7nodes

TGnormal p690 18hr 7nodes

The TG queues are for TeraGrid jobs.

#	Partition	p655	p690
1	interactive	5	1
2	batch	171	0
3	batch	0	7

Name	Nodes	Time limit	Node limit	When
interactive	p655
interactive32	p690
express	p655	2hr	8nodes	weekdays
high	p655	18hr	164nodes
normal	p655	18hr	164nodes
express32	p690	2hr	7nodes	weekdays
high32	p690	18hr	7nodes
normal32	p690	18hr	7nodes
expressL	p690	2hr	1nodes	weekdays
highL	p690	18hr	1nodes
normalL	p690	18hr	1nodes
TGexpress	p690	2hr	7nodes	weekdays
TGhigh	p690	18hr	7nodes
TGnormal	p690	18hr	7nodes

For more information see the NPACI user guide.

Log Format

The original log in available from the NPACI JOBLOG repository, which also includes a description of its format.

Conversion Notes

The converted log is available as SDSC-DS-2004-1.swf. The conversion from the original format to SWF was done subject to the following.

Two lines were omitted due to format problems that prevented reliable parsing.
Interactive jobs are logged without a submit time; the start time was used instead.
The original log specifies the number of nodes each job requested and received. In the conversion this was multiplied by 8 or by 32 to get the number of processors, depending on which type of nodes was used. The node type was gleaned from the queue being *32, *L, or TG*.
The same was done regarding requested processors.
The CPU time is given as a total. This was divided by the number of processors to get the average.
The partition ws also gleaned from the queue being used.
The conversion loses the following data, that cannot be represented in the SWF:
- The "interactive32" queue is mapped to queue 0 together with the "interactive" queue.
The following anomalies were identified in the conversion:
- 149 jobs had undefined end times. In 54 jobs the missing runtime was approximated using the CPU time.
- 4 jobs had 0 start times but a successful completion status. This was changed to -1.
- 22,059 jobs were recorded as using 0 CPU time; this was changed to -1 Of these jobs, 8,976 had "success" status.
- 202 jobs had an average CPU time higher than their runtime. In 149 cases the extra CPU time was larger than 1 minute.
- 10,660 jobs got more runtime than they requested. In 1,044 cases the extra runtime was larger than 1 minute.
- Negative run times occured 87 times; they were changed to 0. All were longer than 1 minute.
- Negative wait times occured 12 times; they were changed to 0. All were longer than 1 minute.

The conversion was done by a log-specific parser in conjunction with a more general converter module.

Usage Notes

The log has a cleaned version available as SDSC-DS-2004-2.1-cln.swf. It is recommended that this version be used.

The cleaning consisted of removing the first 20 jobs, as they seem to represent activity from long before the actual logging started. In particular, the first 10 jobs occur about 3 weeks before the bulk of the work.

The Log in Graphics

File SDSC-DS-2004-2.1-cln.swf

weekly cycle daily cycle burstiness and active users job size and runtime histograms job size vs. runtime scatterplot utilization offered load performance

Parallel Workloads Archive - Logs