1 1
Open Source Cloud Solu.ons Hybrid Cloud: Private with Burst Capacity
Orchestrate and Accelerate Applica.ons
Denis Caromel, CEO Ac.veEon
Cloud computing et Virtualisation : applications au domaine de la Finance
3
ActiveEon Overview
q ActiveEon, a software company born of INRIA, founded in 2007, HQ in the French scientific park Sophia Antipolis
q Co developing with INRIA ProActive Parallel Suite®, a Professional Open Source middleware for parallel, distributed, multi-core computing
q Core mission: Scale Beyond Limits
q Providing a full range of services for ProActive Parallel Suite
q Worldwide production customers and users:
4
A Wide Range of Services
q Training and Certification § Accelerate learning process
q Consulting § Optimize your infrastructure and
maximize ROI q Technical Support - Subscription
§ The guarantee of a quick and efficient assistance
q Integration-Development § Get ActiveEon’s products fine tuned
to your specific needs q Partnerships
§ With OEMs and ISVs
6
ActiveEon accepted in Skolkovo Foundation
6
Moscow International Finance Center, Travaille avec Paris EuroPlace. Techno de l'info, pour la Finance.
9
The ProActive PACA Grid Platform
Production Platform operated by:
Total: q 1 368 Cores (soon 1 428) q 480 GPU CUDA Cores q 30TB Storage (soon 150 TB)
Publically Available Today
11
Technology & Solutions
Cloud & Grid IaaS Physical Machines (Servers, Clusters, Desktops) and Virtual (Hyper-‐V, VMware, KVM, Xen; OpenStack) Dynamic Policies, Full accoun)ng of resource usage Public Cloud Burst (EC2, Azure, Data Centers)
Orchestra.on & Scheduling Mul)-‐Applica)on & Mul)-‐Tenant Portal with Data Management, Remote Visualiza)on APIs (Java, REST, CLI) Integra)on in exis)ng Applica)ons and Web Portals
HPC Workflow & Paralleliza.on Studio for HPC, Workflow Visualiza)on Na)ve Tasks (MPI, OpenMP, Mul)-‐thread, GPU) Java APIs for Paralleliza)on & Distribu)on Matlab & Scilab Distributed Compu)ng
15
ProActive Parallel Suite @ DEXIA, Belfius
• Run Monte Carlo Simulations to predict losses in credit portfolio value due to evolutions in credit parameters
• Compute how much value a portfolio loses/gains in each simulation in order to price a portfolio
• The simulation software runs as a
back-end server and exposes itself as a service addressable by a front-end.
• Deployed on Citrix servers also used for Virtualization of Desktop PCs
Users need reactivity Speed up simulations Distribute the computations
16
ProActive Parallel Suite @ DEXIA, Belfius
Ø 100% Virtual Infrastructure
VM Virtual Machine
Virtual Machine
ProActive Scheduling
ProActive Resourcing
Computing Nodes
User Session
User Session
VM
UC: Acceleration of Financial Valuations
17
C++ library developed by Pricing Partners Pricing solution dedicated to highly complex derivatives,
Greek computation
18
How Does it Work? Price-it Computing Distribution
Price-it Excel
ProActiveScheduler
Pool of sharedresources
Automatic execution via job scheduler
Regular Price-it ExcelInterface
Price-it Excel
19
Accelerated Price-it Performances
Use Case: Bermuda Vanilla, Model American MC
Test conditions: § One computation
is split in 130 tasks that are distributed
§ Each task uses 300ko
Sequential Distributed
More than 3 times faster with only 4 nodes!
4 nodes 5 nodes 6 nodes 7 nodes 8 nodes 9 nodes
Even 6 times faster with 9 nodes!
q Increased Productivity: Reduces Price-it Execution Time by 6 or more!
21
Integration: Scilab and Matlab, Applications
Dedicated resources
LSF
Sta.c Policy
Amazon EC2
EC2
Dynamic Workload Policy
Desktops
Desktops
Timing Policy 12/24
23
OASIS Team Previous work (Doan, 2010)
American/Bermudian basket option pricing in highly distributed & heterogeneous environment (Cloud architecture)
● Use of Monte Carlo pricing methods ● Optimal exercise boundary computation (Ibanez & Zapatero)
● Classification of continuation and exercise values (Picazo)
Use of ProActive for simplified deployment and source
code independence
24
Currently developed (Benguigui, 2012)
High performance computing in hybrid architecture (Cloud/GPU)
q “Optimal” exploitation of GPU (SIMT parallelism, fast access memory use, coalesced memory access, dynamic resource allocation,...)
q Hybrid GPU/Cloud deployment (problem decomposition for fine/coarse grained parallel approach, dynamically/abstractly resource exploitation,..)
q Validation on large-scale parallel platform
Top Related