Download presentation
Presentation is loading. Please wait.
Published byMatilda Williams Modified over 9 years ago
1
Scalarm: Scalable Platform for Data Farming D. Król, Ł. Dutka, M. Wrzeszcz, B. Kryza, R. Słota and J. Kitowski ACC Cyfronet AGH KU KDM, Zakopane, 2013 1
2
Agenda About Us Problem statement Solution Contact 2
3
Who are we ? Computer Systems Group AGH http://www.icsr.agh.edu.pl/ Knowledge in Grids Team http://www.icsr.agh.edu.pl/index.php/knowledge-in-grids-team Close collaboration with ACC Cyfronet AGH 3
4
What are we doing ? Knowledge supported systems for Virtual Organizations Framework for Intelligent Virtual Organizations Grid Organizational Memory X2R Data monitoring and management Service Level Agreement Monitoring QStorMan Self-scalable systems Scalarm 4
5
A short introduction to data farming 5 1.Simulation definition 2. Input space specification 3. Parallel simulation execution 4. Results analysis Process initialization
6
What problem do we address with Scalarm ? Data farming experiments with an exploratory approach Parameter space generation with support of design of experiment methods Accessing heterogeneous computational infrastructure Self-scalability of the management part 6
7
Typical work flow 7 1. Local development 2. A workstation or an institution server 3. National Grid environment 4. Cloud
8
Where is the problem ? 8 1.Integrated Development Environment 2. Shell 3. Shell + scheduler (gLite, QCG...) 4. Shell + proprietary software What about parameter space ? What about fault tolerence? What about collecting results ?
9
What is our solution ? 9 Specify application to run Execute simulation
10
Scalarm overview 10 Self-scalable platform adapting to experiment size and simulation type Exploratory approach for conducting experiments Supporting online analysis of experiment partial results Integrates with clusters, Grids, Clouds
11
Use case – Complete platform for data farming 11
12
Use case – Simulation-as-a-Service 12 API Execute simulation Native app
13
But, does it scale ? 13 Experiment evaluation: Experiment size [#simulations]: 100 000, 200 000, 500 000, 1 000 000, 2 000 000 Computational resources [#servers]: 2, 4, 8, 16 #Clients: 240 * (#servers / 2)
14
Results 14
15
Do you want to now more ? Talk to us during the conference Contact us -> dkrol@agh.edu.pl Visit a website -> http://www.scalarm.com 15
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.