GridOS: Operating System Services for Grid Architectures

1
GridOS Operating System Services for Grid
Architectures
by
Pradeep Padala
2
What is the Grid?

• Electric power grid metaphor
• Ian Fosters 3-point checklist
• Co-ordinates resources that are not subject to
  centralized control
• Using standard, open, general-purpose protocols
  and interfaces
• Delivers nontrivial qualities of service
• A conglomeration of VO (Virtual Organization)s
  coordinating using standard interfaces to solve
  resource intensive problems

3
Grid Jargon

• Middleware
• Globus toolkit (GT) 2nd,3rd versions
• Legion (Avaki)
• UNICORE
• OGSA (Open Grid Services Architecture)
• OGSI (Open Grid Services Infrastructure)
• Various utilities
• globus-url-copy
• Local schedulers (Condor)

4
Why GridOS?

• Tremendous amount of work in middleware
• These middleware use common services and do
  similar things
• Middleware developers re-invent the wheel often
  producing sub-optimal results
• Operating system services for the grid are
  non-existent
• Preliminary work done by PODOS

5
GridOS Design Principles

• Modularity
• Policy Neutrality
• Universality of Infrastructure
• Minimal Core Operating System Changes
• Operating System Independent interfaces and
  protocols

6
Design
7
Modules

• gridos_io (High-Performance I/O Module)
• gridos_comm (Communication Module)
• gridos_rm (Resource Management Module)
• gridos_pm (Process Management Module)

8
High-Performance I/O Module

• Many techniques are available for high
  performance I/O
• But, the techniques are not well incorporated
  into the OS kernel
• Zero-copy procedures
• No copying between user and kernel mode
• No copying between different caches (buffer,
  network, disk) of the kernel
• Buffer tuning
• TCP/IP parameter tuning

9
Communication Module

• Middleware provide similar communication
  primitives to higher layers
• Asynchronous communication
• Synchronous communication
• Parallel transfers

10
Resource Management

• Middleware often have to query local scheduler to
  get resource usage information
• Local Schedulers like Condor use different
  mechanisms on different operating systems to get
  this information
• GridOS provides a common set of resource usage
  statistics tailored for the grid

11
Process Management

• GridOS provides grid-wide process identification
  mechanism
• Unique Ids that can be used by different
  middleware to communicate with each other
• The unique id is made by hostnamepid

12
Implementation

• Implemented in the stable 2.4.19 kernel
  (approximately 4000 lines of code excluding tux
  code)
• Implemented as a patch and a set of modules
• Code from kernel-space web server tux is used

13
Scenario 1 Transporting a file
14
Scenario 2 Copying a file
15
Experiments GridOS vs Proftpd using normal ftp
client
16
Experiments GridOS ftp/ftp client vs
GridFTP/globus-url-copy
17
GridOS ftp/GridOS ftp client vsGridFTP/globus-url
-copy
18
Grid File System
19
Conclusions

• A set of common services in the kernel that
  facilitate grid computing
• Experiments show high performance

20
Documents

• Main thesis page http//www.cise.ufl.edu/ppadala
  /thesis
• Grid File System work
• http//www.cise.ufl.edu/ppadala/grid/gridfs.pdf
• http//www.cise.ufl.edu/ppadala/grid/draft-ggf-gf
  s-gfssurvey-00.pdf
• Questions ?