the open science grid

33
The Open Science Grid NMTIE Conference University of New Mexico, November 2015 Rob Gardner • University of Chicago [email protected]

Upload: rob-gardner

Post on 23-Jan-2017

242 views

Category:

Technology


0 download

TRANSCRIPT

The Open Science Grid

NMTIE Conference University of New Mexico, November 2015

Rob Gardner • University of Chicago [email protected]

Outline

1 32Jumpstart high throughput computing

What training resources are available?

OSG, and how to use it

4campus to national cyberinfrastructure

Jumpstart high throughput computing

1Jumpstart high throughput computing

What is high throughput computing (HTC)?

● Big tasks split into smaller tasks● Batches of similar program runs ● Loops over independent tasks● Many scientific “HPC tasks” are really HTC

4

High throughputmethods

How to bake a very large cake?

Is it necessary to build a very large oven?

5

Q: can the job can be split into smaller tasks?

If so, bake the cake with many small ovens

● use commodity technology

● ovens easier to maintain, replace

● easier to schedule time

6

HTC is many small ovens working on pieces of the problem independently

Why HTC? ⇒ Reduce time to Science

A pattern that can be replicated for many sciences

Lauren MichaelUniversity of WisconsinACI-REF programOSG Summer School 2015

Serial execution Concurrent execution (save time!)7

If the order of subtasks is not important, jobs can be run independently on available processors

Transitioning “serial” to HTC

Lauren MichaelUniversity of WisconsinACI-REF programOSG Summer School 2015

8

Example use case: high energy physics

⇒ Datasets can be processed on independent clusters⇒ Jobs use single or multiple cores on a server

9

Science and computational methods

Any computation that can be split into independent pieces:

● Parameter sweeps● Multi-start simulations● Statistical model

optimization● Image analysis● Pattern recognition● Text mining● Data-intensive analysis

10

● Understanding evolution at molecular scale in DNA with combination of mathematical modeling and simulation

● How quickly does a genome fix a mutation?

● Role of randomness versus natural selection?

Joshua Plotkin, Penn“We use the OSG to run computer simulations for complex processes. By simulating evolution in populations, we can study hypothetical situations that you can’t study in the wet lab or field.”

Example: Evolutionary Biology

Image courtesy Joshua Plotkin. A computationally predicted structure of an influenza virus protein. Plotkin’s research group uses the Open Science Grid to study questions in evolutionary biology and ecology.

11

What is OSG and how can I use it?

2How can users connect to the OSG?

The Open Science Grid

● 500k jobs/day● 925M CPU-

hours/year

13

● A distributed computing partnership of over 125 campuses

● Data intensive: 0.5 PB data transfered/day

Part of an international data-intensive grid

The OSG partners with Europe for global science

14

How do I use it? ⇒ OSG Connect

OSG as a campus cluster

★ Login host★ Job scheduler★ Software★ Storage

15

OSG Connect Service

● http://osgconnect.net site● login node for job management,

login.osgconnect.net

● Stash storage platform — common storage for:○ scp, rsync○ http○ Globus (gridftp)

● Recommended path for (most) new OSG users

16

OSG Connect Service

Has an identity bridge: local campus identity (CILogon) ‣ OSG Connect identity (Globus) ‣ virtual organization (OSG)+ HTCondor Glidein Overlay

⇒ Goal is to provide a virtual HTC cluster experience

17

Software & tools on the OSG

● Distributed software file system OASIS● Special module command

○ identical software on all clusters○ 170 libraries

#!/bin/bash

switchmodules oasismodule load R

module load matlab

...18

Storage service: “Stash”

● Storage service for job input/output data● Globus Server for managed transfers between

campus and the OSG ● POSIX access provided to the login host● 500 TB capacity● Personalized http accessible space● Connected to 100 Gbps SciDMZ (I2, ESnet)

19

Simulations for high rate error correction codes for optical data communication and data compression

● David Mitchell, NMSU EE faculty○ Ahmad Golmohammadi (EE graduate student) ○ Important for digital space and satellite

communication & wireless data transmission○ Whole system simulations - transmitter, decoder,

receiver & stochastic noise, data compression○ Computations are well-suited to HTC. Ahmad:

Local contact: Piyasat Nilkaew, Dir. Telecom, Networking, User Support

The OSG Connect Client:bring OSG pools to Campus

21

OSG Connect - onboard quickly

Researchers without local HPC or HTC can login to OSG Connect directly,but….

22

Bring the submit point to campus?

...some researchers prefer to work “from home”:● Local standard

configuration● Local standard data

management● Local standard

software access tools● First point of contact

for scientific computing consulting

23

Connect Client brings pools to campus

Idea is to bring the submit point “home” i.e. on campus

Submit locally,run globally

24

Training Resources

3What training resources are available?

Annual OSG User School (July, Madison)

Week of lectures and intensive HTC challengesTargeted for grad, postdocs & HPC facilitators

Joint OSG Software Carpentry Workshops

● Two members of OSG User Support team are Software Carpentry instructors

● Extend standard core curriculum with a day of HTC computing best practices

HTC Recipes, Online Help, Ticketing

● Helpdesk portal● Knowledge

base● Online chat● Submit help

requests

Connecting campuses to national cyberinfrastructure

29

4campus to national cyberinfrastructure

How can I get my campus connected?

● Connect your campus researchers to the OSG○ OSG Connect - a login service to the OSG○ Connect Client : a job submit client for the local cluster

■ provide a “burst” like capability for HTC jobs to shared opportunistic resources

● Connect a campus cluster to OSG○ Lightweight connect : OSG sends “glidein” jobs to your cluster, using a

simple user account■ No local software or services needed!

○ Large scale: deploy the OSG software stack■ Support more science communities at larger scale■ Usually recommend as a second step

30

31

Campus to national networking

● OSG helps campuses analyze PerfSonar data● Can create campus or regional dashboards

32

Summary of the Open Science Grid

● Helps researchers speed up their research using high throughput computing methods

● Helps campus HPC administrators share local resources for multi-campus and national collaborative research

● Collects, archives and analyzes links between campuses and the national cyberinfrastructure

osgusers

[email protected]

support.opensciencegrid.org

www.opensciencegrid.org/links

opensciencegrid

33This talk:http://bit.ly/1WZ2grg