Managing Disaster Response through On

Managing Disaster Response through
On‐Demand Resource Federation
Dr. Craig A. Lee, Senior Scientist, [email protected]
Dr. Nehal Desai, Senior Engineering Specialist
Andrew Brethorst, Member Technical Staff
The Aerospace Corporation
(a California nonprofit corporation that operates a
federally funded research and development center)
© The Aerospace Corporation 2014
The GeoINT Community Cloud Demo
2
Used by permission of the NCOIC
What Did We Learn From All This?
• While managing the sharing access to storage containers
is a good first step, it was inadequate to facilitate the
secure federation of arbitrary tools, i.e., arbitrary services
•
Managing federated access to databases and RSS feeds could
have been used by GCC participants
• Federation management needed to be generalized to
manage any type of service endpoint
•
Cloud infrastructure services (VMs, storage, SDNs) are just
services, just like application-level services
• How can the VO concept be generalized to manage
general federations?
3
Virtual Organization Abstract Concepts
•
A VO is a security and collaboration context not exclusively associated with any one physical organization or site
– Participating partners agree upon structure, rules and processes
– A VO partner can be a single person, a group or an entire organization
•
A VO has members that are assigned roles and/or attributes
– Membership roles or attributes grant specific capabilities within a given VO as determined by each resource/service provider
•
Partners participating in a VO contribute resources, i.e., data and services
– They retain complete control over their own resources!
– Access by VO members can be modified or revoked at any time by both the VO administrator and the resource administrator
•
A VO Management System (VOMS):
– Maintains member identity attributes and authorization attributes
– Enables resource (service) discovery
– Enables validation of VO member authz credentials on service invocation
•
Many implementation options and choices necessary
– Centralized (third‐party) and Distributed (P2P) implementations possible
4
General AuthN/Z
1
UserA
2
3
IdPA
5
4
6
SPA
5
AuthN/Z in a Distributed Env
1
1
UserA
2
3
IdPA IdPB
5
4 4
5
3
6
6
SPA SPB
Request to SPB with IdPA credentials?
6
UserB
2
Federated AuthN/Z Using a VOMS
VOMS
1
1
UserA
2
3
IdPA
5
4
IdPB
4
5
3
6
6
SPA
7
UserB
2
SPB
General Federation Mgmt Design Requirements
•
•
•
•
•
•
Client-Side: Federated Authentication & Service Discovery
Server-Side: Federated Credential Validation & Authorization
Centralized, Trusted, Third-Party VOMS
Proxy VOMS
Distributed, Trusted, P2P VOMS
Trust Federations
Diagram illustrates an
external, third-party VOMS.
8
KeyVOMS: A Keystone-based VOMS
• KeyVOMS is a centralized, third-party VOMS
• Re-purposed instance of the OpenStack Keystone v3 service
• Keystone maintains a Service Catalog
• Used in KeyVOMS as the VO Services Catalog for service discovery
• Keystone v3 Object Model very close to what’s needed
Horizon
OpenStack
Core Services
Nova
Glance
Keystone
9
Swift
Current Keystone v3 Object Model
with representative object associations (assignments)
…
Role_0
Role_j-1
Services &
Endpoints
Domain
Users
Svc_0
Endp
User_0
…
User_i-1
Projects
… Endp
…
Roles
Proj_0
…
Proj_k-1
10
Svc_m-1
Endp
… Endp