David, Joel,
Need a little help from the Standards/Technical team here… In the process of writing the Grid and Distributed Computing Landscape document, David Birnbaum has come up with a series of questions that need to be answered for the reader. Can you look these over and do the following: (a) Answer these questions from your perspective; (b) identify an OGF expert who might be able to answer one or more of these questions and send them the question and solicit them to provide a response. The responses can be in the form of a short paragraph per question but try to be as specific and factual as possible in your answer. David will utilize your responses and others as in-depth source material as he continues to write the document.
QUESTIONS: 1. What are the issues or limitations of a virtualized server infrastructure that can be addressed by grid capabilities? Virtualizing a server infrastructure only addresses one layer in the system. The Grid provides, using many of the same concepts as SOA, virtualization of higher layers of the stack, e.g. application etc. 2. What are the issues or limitations that might be encountered during either a pilot or production implementation of SOA that can be addressed by grid capabilities? The key issues here are scale and dynamics. Most SOA deployments are relative small scale of highly specialized. Adding Grid to SOA
3. Applications a. Which non-compute intensive applications are well suited to running on a grid? This one is out of my scope. b. What modifications are needed to run an application on a grid? Does the application have to support parallel processing? If the aim of using Grid is to exploit many servers on a single task,
[Bicycle crash boiler plate: I will include this boiler plate in my mails for a few weeks to apologize for terseness in the rest of the mail. I have a broken finger from the crash on Oct 4th, and typing is hard and possible only in short spurts. Sorry, Dave.] Folks on the TSC, Mark needs a little help, and on a short fuse. I have tried to outline some simple ideas to these questions, but I think David B. needs more detail than I can provide at present. On 21 Nov 2006, at 18:38, Linesch, Mark wrote: provides a more dynamic platform, allowing new service to come and go continuously. then yes parallelism is necessary. This does not necessarily mean the application needs adaptation. Many applications are already parallel in nature or the parallelism can be supported by parts of the Grid middle ware. For example, most transaction based applications would not require changing, provided that the underlying database provided support for distributed transactions.
c. Are I/O intensive applications not suitable for grid? Provided that the infrastructure and application are designed for it, data intensive applications work very well on a Grid. care is needed in ensuring that data transfer bottle necks are managed and that the data itself is distributed across the Grid infrastructure. The applications should be configured to tolerate latency, e.g. using work queues. 4. For an enterprise IT architect, what grid standards, services and technologies does OGF recommend that they have on their planning horizon over the next three years? The list in the Technical Strategy Document is a good start. I would add to that standards for data centre management, which may come from OGF or elsewhere.
Please email your perspective/response to these questions directly to David and copy me by EOD, Wed, November 29th. Your help is much appreciated. I know this is less text than you need, but at least it's on time. I hope the rest of the TSC can help.
-- Take care: Dr. David Snelling < David . Snelling . UK . Fujitsu . com > Fujitsu Laboratories of Europe Hayes Park Central Hayes End Road Hayes, Middlesex UB4 8FE +44-208-606-4649 (Office) +44-208-606-4539 (Fax) +44-7768-807526 (Mobile)
My $0.02USD Tom _______________________________________________
Folks on the TSC,
Mark needs a little help, and on a short fuse.
I have tried to outline some simple ideas to these questions, but I think David B. needs more detail than I can provide at present.
On 21 Nov 2006, at 18:38, Linesch, Mark wrote:
David, Joel,
Need a little help from the Standards/Technical team here. In the process of writing the Grid and Distributed Computing Landscape document, David Birnbaum has come up with a series of questions that need to be answered for the reader. Can you look these over and do the following: (a) Answer these questions from your perspective; (b) identify an OGF expert who might be able to answer one or more of these questions and send them the question and solicit them to provide a response. The responses can be in the form of a short paragraph per question but try to be as specific and factual as possible in your answer. David will utilize your responses and others as in-depth source material as he continues to write the document.
QUESTIONS: 1. What are the issues or limitations of a virtualized server infrastructure that can be addressed by grid capabilities? Virtualizing a server infrastructure only addresses one layer in the system. The Grid provides, using many of the same concepts as SOA, virtualization of higher layers of the stack, e.g. application etc.
2. What are the issues or limitations that might be encountered during either a pilot or production implementation of SOA that can be addressed by grid capabilities? The key issues here are scale and dynamics. Most SOA deployments are relative small scale of highly specialized. Adding Grid to SOA provides a more dynamic platform, allowing new service to come and go continuously. 3. Applications a. Which non-compute intensive applications are well suited to running on a grid? This one is out of my scope.
Business intelligence/analytics. These applications while actually 'compute' intensive are not what most consider w.r.t. that term. ETL processes are also well suited to grids.
b. What modifications are needed to run an application on a grid? Does the application have to support parallel processing? If the aim of using Grid is to exploit many servers on a single task, then yes parallelism is necessary. This does not necessarily mean the application needs adaptation. Many applications are already parallel in nature or the parallelism can be supported by parts of the Grid middle ware. For example, most transaction based applications would not require changing, provided that the underlying database provided support for distributed transactions.
Absolutely agree, in fact most 3-tier applications are incredibly well suited to run on a grid. The noticeable exception is the final tier (database). Only some vendors have database assets that work well on a grid. Even those have difficulty with the dynamic provisioning aspects of grid infrastructure.
c. Are I/O intensive applications not suitable for grid? Provided that the infrastructure and application are designed for it, data intensive applications work very well on a Grid. care is needed in ensuring that data transfer bottle necks are managed and that the data itself is distributed across the Grid infrastructure. The applications should be configured to tolerate latency, e.g. using work queues.
Not sure I agree with the last statement. There are many new techniques for building highly scaleable data grid infrastructures absent latency. The real question is access to the data parallel or serial and that is an application design issue.
4. For an enterprise IT architect, what grid standards, services and technologies does OGF recommend that they have on their planning horizon over the next three years? The list in the Technical Strategy Document is a good start. I would add to that standards for data centre management, which may come from OGF or elsewhere.
I would add data center fabric standards, semantic web standards, storage virtualization standards off the top of my head.
Please email your perspective/response to these questions directly to David and copy me by EOD, Wed, November 29th. Your help is much appreciated. I know this is less text than you need, but at least it's on time. I hope the rest of the TSC can help.
-- Take care: Dr. David Snelling < David . Snelling . UK . Fujitsu . com > Fujitsu Laboratories of Europe Hayes Park Central Hayes End Road Hayes, Middlesex UB4 8FE +44-208-606-4649 (Office) +44-208-606-4539 (Fax) +44-7768-807526 (Mobile) _______________________________________________ tsc mailing list tsc@ogf.org http://www.ogf.org/mailman/listinfo/tsc
3. Applications a. Which non-compute intensive applications are well suited to running on a grid?
There's a wide range of transactional, three-tier applications that are response-time sensitive, that are well suited to running on a grid - where the grid infrastructure's flexibility and dynamic reconfigurability can be brought to bear. For example, a "web store" can use grid technologies to repurpose the servers in a pool to respond to ebbs and flows in customer demand, both at the HTTP-serving leaves and the middleware engine layers. Grid-enabled database servers and storage units can similarly respond to increased demand to spread the load of additional customers or quarter-end tasks. When the load subsides, the storage can be deprovisioned and put to use elsewhere.
c. Are I/O intensive applications not suitable for grid?
With a suitable storage infrastructure, I/O intensive applications can be supported by grid-enabled servers and storage structures that spread the load as needed to provide the promised or requested response time ... at a price, of course!
Both of these questions seem to have an implicit assumption that we are talking about mapping applications to "idle desktops" or some similar environment with poor reliability and I/O capability. As Chris points out, grid technologies can also be used to manage pools of highly reliable and well-connected servers. Perhaps as well as answering the questions as below, we should address the misunderstanding that seems to be inherent in the question. For (a), as a further example to that specified by Chris, we could be dynamically allocating CPUs used to host virtual desktops. With respect to (c), Ioan Raicu is doing some nice work in which he greatly *increases* I/O capability using grid techniques, by dynamically allocating an increasing number of servers to a data analysis workload as the amount of work increases. Chris Kantarjiev wrote:
3. Applications a. Which non-compute intensive applications are well suited to running on a grid?
There's a wide range of transactional, three-tier applications that are response-time sensitive, that are well suited to running on a grid - where the grid infrastructure's flexibility and dynamic reconfigurability can be brought to bear.
For example, a "web store" can use grid technologies to repurpose the servers in a pool to respond to ebbs and flows in customer demand, both at the HTTP-serving leaves and the middleware engine layers.
Grid-enabled database servers and storage units can similarly respond to increased demand to spread the load of additional customers or quarter-end tasks. When the load subsides, the storage can be deprovisioned and put to use elsewhere.
c. Are I/O intensive applications not suitable for grid?
With a suitable storage infrastructure, I/O intensive applications can be supported by grid-enabled servers and storage structures that spread the load as needed to provide the promised or requested response time ... at a price, of course!
_______________________________________________ tsc mailing list tsc@ogf.org http://www.ogf.org/mailman/listinfo/tsc
-- Ian Foster, Director, Computation Institute Argonne National Laboratory & University of Chicago Argonne: MCS/221, 9700 S. Cass Ave, Argonne, IL 60439 Chicago: Rm 405, 5640 S. Ellis Ave, Chicago, IL 60637 Tel: +1 630 252 4619. Web: www.ci.uchicago.edu. Globus Alliance: www.globus.org.
participants (4)
-
Chris Kantarjiev -
David Snelling -
Ian Foster -
Maguire_Tom＠emc.com