GriT: A CORBA-based Grid Middleware Architecture

1
GriT A CORBA-based Grid Middleware Architecture
Presented by Dr. Jeff Gray Aniruddha Gokhale,
Balachandran Natarajan gokhale,
bala_at_isis-server.isis.vanderbilt.edu ISIS,
Vanderbilt University www.isis.vanderbilt.edu/gok
hale/PDF/HICSS-GRID03.pdf
2
Motivation

• Grid Application Characteristics
• Multiple end-to-end QoS properties
• CPU and data intensive (petaflops, petabytes)
• Secure, controlled access to resources from
  multiple service providers gt need individual
  SLAs
• E.g., collaborative scientific applications,
  virtual surgery

3
Grid Application Development Challenges
Standards-based programmable middleware enhanced
with Web API
1. Tight coupling with COTS grid infrastructure
middleware (GIMs)
2. Proliferation of GIMs, such as Globus, ICENI,
Legion
3. Need to satisfy simultaneous multiple QoS
4. No single sign-on secure access capability to
resources
5. Accidental complexities assembling and
deploying grid applns
6. Lack of resource reservation QoS
provisioning patterns
4
Grid TAO (GriT) Architecture

• Based on principles from DP-CORBA, RT-CORBA, CCM
• Seamless integration with GIMs via extensible
  transport capabilities
• Uses web-based interfaces for service
  provisioning
• Implemented using TAO and CIAO technology

5
Grid Service Provider (GSP)

• Web service access to underlying middleware
  (e.g., OGSA)
• Single sign-on capabilities
• GSP maintains individual SLAs on behalf of user
• Collaborations via session initiation protocol
  (SIP)

6
Meta-Resource Broker (MRB)

• Broker for resources from underlying service
  providers
• Maintains abstractions of concrete resources
• Uses DP-CORBA parallel and part objects for
  parallel resource reservation

7
MRB Part Object Core

• Uses TAOs pluggable protocol framework to talk
  Grid protocols
• Used for resource discovery and reservation

8
Avoiding Starvation of Transient Applns

• Context peer-to-peer collaborative applns will
  require resources for short durations
• Problem persistent long duration applications
  might starve short duration applications
• Solution GriT uses slack stealing algorithms
  by diverting resources from persistent
  applications to short duration applications
  during periods of non-use but returns them to
  persistent applns whenever required.

9
Transparent Dispatching of Requests

• Context Incoming appln request to MRB is
  received by the virtual object representing
  collection of reserved resources
• Problem Incoming requests to MRB must be
  transparently fanned out to all the required
  resources
• Solution GriT enhances the Portable Object
  Adapter to maintain actual address of the
  resources so that incoming requests are
  transparently forwarded to desired resource.

10
Ensuring Dependability

• Context Dependability is key QoS requirement
  for Grid applications
• Problem The collection of resources maintained
  by the MRB must now be a collection of redundant
  resources that needs a policy to determine the
  order in which these resources are used and
  minimizing overprovisioning
• Solution GriT employs TAOs pluggable load
  balancing service that allows service providers
  to insert custom strategies for minimizing
  overprovisioning

11
Interacting with Grid Protocols

• Context Grid protocols such as Globus are
  typically categorized into 4 layers
• Problem There is no standardization yet on how
  the underlying grid protocols are structured.
  Existing applns are tightly coupled to these
  protocols.
• Solution GriT uses TAOs pluggable transport
  bridge solution to hook up to any underlying
  transport.

12
Simplifying use of GriT

• Context Both wireless and wireline clients need
  to participate in collaborative grid applns
  requiring a thin client solution
• Problem Programming at the GriT level might
  also tightly couple the client with GriT API
  while also increase footprint
• Solution GriT services are made available as a
  web service. This approach is in tune with the
  emerging Open Grid Services Architecture approach

13
Summary of Ongoing Work

• FT and RT capabilities in TAO done
• DP-CORBA implementation on going
• CIAO CCM implementation at Washington Univ DOC
• Using model integrated computing to automate
  provisioning tasks
• Testing to be done using Vanderbilts Vampire
  cluster and Utahs Emulab

www.isis.vanderbilt.edu
14
Concluding Remarks

• GriT enhances TAO/CIAO
• Meta resource broker
• Patterns for resource reservation and QoS
  adaptation
• Uses Model Integrated Computing tools
• GriT will be available at
• www.isis.vanderbilt.edu