OSDF Cache Selection

April 4th, 2025

Fabio Andrijauskas - University of California San Diego

The OSPool leverages the Open Science Data Federation (OSDF) to efficiently provide data access for computing jobs and software efficiently, ensuring seamless and reliable data transfers. One of its most outstanding features is its ability to dynamically locate the closest cache for each data transfer, minimizing latency and improving overall performance. By optimizing data movement, the OSPool enhances accessibility to large datasets, reduces network congestion, and accelerates computational workflows. This distributed approach ensures that researchers and scientists can efficiently access the data they need, regardless of their physical location, enabling faster and more effective scientific discoveries.

Executive summary

This document presents the results of testing the OSDF cache selection mechanism, which was exercised by repeatedly requesting files from most of the geographic locations that serve the OSPool. Most of the tests were positive, delivering data from the expected cache location, but the OSDF cache selection did pick a suboptimal cache at a couple of geographical locations. The behavior was consistent in both low-load and in stress test setups.

Recommendations

There is one recommendation for the Pelican Platform Team:

· Create a tool to check the site and the selected cache periodically. This tool can be used to measure the transfer rate from sites and caches.

Next, the recommendations for the OSDF operations are:

· Check why the Pelican Platform can’t reach the proper cache in very few cases.

· Check the GeoIP positions for the sites that got the wrong closest cache.

Finally, the recommendations for the OSPool operations are:

· Create a pattern for OSPool site and node names (e.g., Country_State_City_baseInternetDomain_Project).

Detailed explanation

The tests were performed using ap21.uc.osg-htc.org and requested a list for all available sites in the OSPool using

condor_status -pool gfactory-1.osg-htc.org -any -const 'MyType=="glidefactory"' -af GLIDEIN_Supported_VOs GLIDEIN_Site | grep OSGVO | sort -u | awk {'print $2'} | uniq > sites

and sites from the K6 testing tool. Each site was tested 15 times using one script using HTTP requests to the OSDF director and using the pelican client; the methodology was based on checking the site location and cache selected by the OSDF Director. Table 1 shows the name, location, and the location of the cache for the easy test. Furthermore, a stress test was executed on each site, requesting 100 files in parallel and checking the location.

  • Sixty-eight sites were used in the test; only four sites informed a wrong closest cache. Considering only OSG sites, only two sites showed the wrong location.
  • Two sites informed different positions using the Pelican client and the HTTP request.
  • The stress test did not show any wrong locations for the transfers.
  • A different cache was reported whenever the closest cache location was wrong.


Table 1: Site name, cache selected, and actual site location. The ‘k6 site’ is a host used for the K6 load testing tool; they use Amazon Cloud hosts.

Site name

Selected Cache and Location/State

Site Location/State

Closest location -Pelican director

Closest location -pelican client

singapore k6 site

singapore.nationalresearchplatform.org - Singapore

Singapore

yes

-

brazil k6 site

osdf-cache.sprace.org.br - Brazil

Brazil

yes

-

tiger-osg-backfill-prod

osdf-uw-cache.svc.osg-htc.org - Wisconsin

Wisconsin

yes

yes

clemson.edu

dtn-pas.cinc.nrp.internet2.edu - Ohio

South Carolina

yes

yes

empest-epyc

dtn-pas.bois.nrp.internet2.edu - Idaho

MT

yes

yes

lp126.lonepeak.peaks

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

mortimer.hpc.uwm.edu

dtn-pas.bois.nrp.internet2.edu - Idaho

Wisconsin

yes

yes

chpc.utah.edu

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

int.chpc.utah.edu

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

arkansas state university

unl-cache.nationalresearchplatform.org - Nebraska

Arkansas

yes

yes

amazon:us:columbus k6 site

osg-new-york-stashcache.nrp.internet2.edu – New York

Ohio

yes

-

amazon:ca:montreal k6 site

mghpcc-cache.nationalresearchplatform.org - Massachusetts

Montreal -Canada

yes

-

amazon:de:frankfurt k6 site

amst-fiona.nationalresearchplatform.org – Amestrdan - Netherlands

Frankfurt - Germany

yes

-

amazon:gb:london k6 site

lond-osdf-xcache01.es.net – London - UK

London - England

yes

-

amazon:fr:paris k6 site

lond-osdf-xcache01.es.net– London - UK

Paris - France

yes

-

amazon:us:palo alto k6 site

dtn-pas.cinc.nrp.internet2.edu – Ohio | dtn-pas.bois.nrp.internet2.edu - Idaho

California

no

-

amazon:us:portland k6 site

osg-chicago-stashcache.nrp.internet2.edu – Chicago | mghpcc-cache.nationalresearchplatform.org – Massachusetts

Portland

no

-

amazon:us:ashburn k6 site

osg-new-york-stashcache.nrp.internet2.edu – New York

Virginia

yes

-

alabama-chpc

dtn-pas.jack.nrp.internet2.edu - Florida

Alabama

yes

yes

amnh

osg-new-york-stashcache.nrp.internet2.edu – New York

New York

yes

yes

asu-sol

fdp-d3d-cache.nationalresearchplatform.org - California

ncar-cache.nationalresearchplatform.org Colorado

Arizona

no

no

beocat-slate

unl-cache.nationalresearchplatform.org -Nebraska

Kansas

yes

yes

cameron university

unl-cache.nationalresearchplatform.org -Nebraska

Oklahoma

yes

yes

center for advanced research computing

fdp-d3d-cache.nationalresearchplatform.org - California

California

yes

yes

chtc

osdf-uw-cache.svc.osg-htc.org - Winscosin

Winscosin

yes

yes

chtc-spark

osdf-uw-cache.svc.osg-htc.org - Winscosin

Winscosin

yes

yes

colorado

osg-new-york-stashcache.nrp.internet2.edu director. – New York
osg-houston-stashcache.nrp.internet2.edu pelican - Texas

Colorado

no

no

duke-ncshare

dtn-pas.cinc.nrp.internet2.edu - Ohio

North Calorina

yes

yes

elsa

osg-new-york-stashcache.nrp.internet2.edu – New York

New Jersey

yes

yes

fandm-its

osg-new-york-stashcache.nrp.internet2.edu – New York

Philadelphia

yes

yes

fnal

osg-new-york-stashcache.nrp.internet2.edu – New York
fdp-d3d-cache.nationalresearchplatform.org – California

Illinois

yes

yes

fnal_gpgrid

dtn-pas.bois.nrp.internet2.edu director - Idaho
ncar-cache.nationalresearchplatform.org pelican – Colorado

Illinois

no

no

gatech

dtn-pas.jack.nrp.internet2.edu - Florida

Gerogia

yes

yes

grid_ce2

osg-chicago-stashcache.nrp.internet2.edu - Michigan

Michigan

yes

yes

gsu-acids

dtn-pas.jack.nrp.internet2.edu - Florida

Georgia

yes

yes

hawaii-koa

osg-sunnyvale-stashcache.nrp.internet2.edu - California

Hawaii

yes

yes

iitisi

osdf-uw-cache.svc.osg-htc.org - Winscosin

Michigan

yes

yes

langston-lion

osg-houston-stashcache.nrp.internet2.edu - Texas

Oklahoma

yes

yes

lsuhsc-tigerfish

dtn-pas.hous.nrp.internet2.edu - Texas

Louisiana

yes

yes

maine-acgmaine-penobscot

mghpcc-cache.nationalresearchplatform.org – Massachusetts

Maine

yes

yes

michigan

osg-chicago-stashcache.nrp.internet2.edu - Illinois

Michigan

yes

yes

mi-horus

dtn-pas.bois.nrp.internet2.edu - Idaho

Missouri

yes

yes

missouri-hellbender

osg-chicago-stashcache.nrp.internet2.edu - Illinois

Missouri

yes

yes

mtstate-tempest

dtn-pas.bois.nrp.internet2.edu - Idaho

Montana

yes

yes

numepodu-ubuntuoru-titan

unl-cache.nationalresearchplatform.org - Nebraska

Oklahoma

yes

yes

osg_us_fsu_hnpgrid

dtn-pas.jack.nrp.internet2.edu - Florida

Florida

yes

yes

pdx-coeus

dtn-pas.bois.nrp.internet2.edu - Idaho

Portland

yes

yes

psu-ligo

osg-new-york-stashcache.nrp.internet2.edu – New York

PA

yes

yes

puertorico

dtn-pas.jack.nrp.internet2.edu - Florida

Puerto Rico

yes

yes

purdue-anvil

dtn-pas.bois.nrp.internet2.edu - Idaho

Purdue

yes

yes

rhodes-hpc

dtn-pas.cinc.nrp.internet2.edu - Ohio

Tenesse

yes

yes

siue-cc-production

osg-chicago-stashcache.nrp.internet2.edu - Illinois

Illinios

yes

yes

swan

unl-cache.nationalresearchplatform.org - Nebraska

Nebraska

yes

yes

swarthmore-firebird:

osg-new-york-stashcache.nrp.internet2.edu – New Work

Pennsylvania

yes

yes

tntech-warp1

dtn-pas.cinc.nrp.internet2.edu - Ohio

Tenesse

yes

yes

denveruchicago

dtn-pas.denv.nrp.internet2.edu - Colorador

Denver

yes

yes

ucmerced-pinnacles

osg-sunnyvale-stashcache.nrp.internet2.edu - California

California

yes

yes

uconn

mghpcc-cache.nationalresearchplatform.org - Massachusetts

New York

yes

yes

uconn-hpc

mghpcc-cache.nationalresearchplatform.org - Massachusetts

New York

yes

yes

unespunr-cc

osg-sunnyvale-stashcache.nrp.internet2.edu - California

Brazil

no

no

usd-lawrence

unl-cache.nationalresearchplatform.org - Nebrasja

South Dakota

yes

yes

utah-granite

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

utah-kingspeak

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

utah-lonepeak

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

utah-notchpeak

dtn-pas.bois.nrp.internet2.edu - Idaho

Utah

yes

yes

uw-it

unl-cache.nationalresearchplatform.org - Nebraska

UW

yes

yes

uwm-mortimer

osdf-uw-cache.svc.osg-htc.org - Wisconsin

Wisconsin

yes

yes

emporia state university

unl-cache.nationalresearchplatform.org - Nebraska

Kansas

yes

yes