D and Domain (Logical & Physcial)

S10 has physical domain control data structure and physical domain control data structure are defined for platform dependent sun4u architecture. From T2000, logical domain will be introduced with new firmware on virtualization. DCS

May I ask a few questions below:

(1) it seems common psinfo_t and cpuinfo_t does address platform specific
information. What is the proper way to access the domain realted structure
so that I can lookup related cpuset and domain state or domain heartbeat ?

(2) how D structure or uts structure provide the visiability to understand if the process is running on physical or logical domain ?

(3) how domain related kernel resource usage other then (1) and (2) will be handled ?

Zone and Domain (Physical & Logical)

With the enhancement of solaris virtualization and partition model
May I ask if and how to approach the things I am working on.

(1) S10 does have platform specific sun4u domain data and domain control
data structure with abstracted domain state and domain heart beat etc.
how to lookup zone is associated with which physical domain and proc
running within a specific zone associated with domain ?

(2) how to make the difference that a proc is running within which
physical domain or logical domain ?

(3) how to lookup domain and zone specific resource usage ?

Grid based MMP games

(1) MMP games via virtual on-demand service
(2) game developers -- performance and flexiability
developer want fine control the resources over
grid ---- Grid is developer platform for with
resource management and provisioning
(3) game publishers -- multiple games in one resource base
(4) service providers --- distribute the game state
and associate load to multiple grid nodes
(5) Service provider hosts the games closet to
the geo where player in game so that do not
need to route traffic over the net

Decentralization introduces its own problems. Game
states must be kept consistent across servers.

The key is the services and protocols to mainitainn
the distributed game state


1. Client side render the data translated into inherent
protocols by middle layer
2. UDP/IP thin layer on transport
3. gateway is single entry point to a section or region
games
4. game server host all game rules and logics.
6. Data Store for persistence
7. Grid service for resource management

CMT RSA

CMT cores can be dynamically disabled in the
case of core failure and Solaris can continue to run with LWPs
rescheduled on remaining cores!

So, in addition to all the reliability due to simplicity arguments,
the lower heat arguments, the ECC/parity checking of L1/L2 cache/
main DRAM, and the thermal management RAS features, T1 does have a
Defect Engine (DE) that will blacklist individual threads or a core
(depending upon what FMA associates the failure with) and Solaris will
continue to run on remaining cores.

Of course there can be certain hard errors that force the system to
reboot with blacklisted threads/cores, but that can also be the case
with UltraSPARC III/IV.

The L2 cache could perhaps be considered a single point of failure, but
L2 cache does have Error Correcting Code protection so is no less fault
tolerant than most RAID setups, but with far longer MTBF. In the event
of kernel access that gets an uncorrectable L2 cache error on a dirty
line (note the 3 unlikely criteria that much each be met), the system
will crash, but I'm sure there are similar edge cases for most SMP
platforms.

Overall Kabira KTS on T1 on T2000 should be an extremely reliable and
fault tolerant CPU and platform!

Unix OS minor

(1) On Unix, command interpreter-- shells
(2) On unix, command interpreter does not know the command, it uses the command to indentify the file to be loaded into memory and executed. Thus the Unix command to delete a file:

rm file.txt

would search for a file called "rm", load the file into memory, and execute it with the parameter
"file.txt" The function associated with the "rm" command would be definied completely by the code
in the file rm. In this way, new files with the proper name can be easily handled

Top Six in Grid 2006

[ ] M540387 ) Grid's Top Six in 2006
By Greg Nawrocki, Contributing Author

Globus Consortium president Greg Nawrocki takes a look at the "Top Six
Grid Areas to Watch in 2006" and the "Six Interesting Vendors to Watch
in Enterprise Grid in 2006." Among the areas covered are "Securing the
Grid" and "The Linux/Grid Relationship"; vendors discussed include
Platform, EMC and Microsoft.

---

**TOP 6 GRID AREAS TO WATCH IN 2006**

Data Virtualization and Grid Computing Directions

One of the primary strengths of e-Science Grids has historically been
their ability to transfer large data sets for collaborative computing
scenarios. And enterprise today is similarly moving beyond the
traditional "compute grids" and starting to explore interests in the
potential of "data grids" for data-intensive scenarios such as
business intelligence and handling the movement and management of
transactional data. While experts such as Carl Kesselman warn us not
to over-taxonomize or create a false sense of mutual exclusivity
between data and compute grids (after all, he says, "you can't do much
computing without data, and you generally don't have much data without
doing compute"), the confluence of data virtualization and Grid
computing trends in enterprise should be interesting to watch in '06.

Securing the Grid

With the release of Globus Toolkit 4.0 -- the convergence between Grid
services and Web services took another giant step forward. But Grid
services do have a number of security challenges that are sometimes
greater (and sometimes different) than security challenges for Web
services. As users cross virtual organizations, grids need mechanisms
to accommodate all of the different security models and frameworks.
The WS standards work is certainly a key to getting over the
interoperability hump between different security standards, but grids
also require very fine-grained ways to express delegations of rights
(who gets access to which data and compute resources), standard ways
for Grid services to navigate firewalls, and more "enterprise-ready"
mechanisms for paper trails and auditing. Security will continue to be
a very busy focus area for Grid developers in '06.

Network Intelligence Increasing the Grid's IQ

Networking is not just about "pipes" and "plumbing" anymore, it's much
smarter than that. There are a number of networking heavies that are
introducing new virtualized network control and access to the Grid --
from time-of-day reservation scheduling capabilities, to
"application-aware" monitoring, to smarter networking devices that can
more effectively route Grid traffic. As Bob Aiken from Cisco said in
'05, what we're seeing on the networking side in grids today is "a
blurring of the boundaries between operating systems, networks and
middleware." In 2006, Grid pros will start thinking of the network
more like neurons in a biological sense and less like pipes in a
plumbing sense.

The Linux/Grid Relationship

Today's largest IT shops, including powerhouses like Google, Yahoo and
Amazon, are running "scale-out" Linux environments. While the key to
Grids is enabling heterogeneity and making it so that it "doesn't
matter what your OS is," the openness, low cost and flexibility of
Linux has made it the preferred OS for Grid computing. Over the last
couple of years, Linux virtualization has seen a huge boost in support
from the major distributions -- which catapulted virtualization into
mainstream enterprise. What will the Linux Grid computing efforts look
like in 2006, and will they rally behind de facto middleware like the
Globus Toolkit for Grid the same way that they rallied behind Xen for
server virtualization?

Grid Licensing Issues

In 2005, William Fellows from The 451 Group made a compelling case
(link to abstract of report) that enterprise applications' traditional
licensing schemes are fundamentally at odds with how applications are
consumed in Grid environments. Reminiscent of the SCO licensing fears
in the history of Linux adoption, some enterprises question whether
Grid application licensing issues have been sufficiently addressed.
Perhaps 2006 will be the year that an SAP or other large application
vendor takes a definitive stand on what licensing will/should look
like in enterprise Grid computing environments.

Virtual Workspaces

To understand the importance of virtual workspaces in the Grid, it
helps to paraphrase one of the leading researchers on the subject,
Kate Keahey of Argonne National Laboratory
(, who
says, "A virtual workspace is an abstraction of an execution
environment that can be made dynamically available in the Grid. One
requirement of a virtual workspace is the ability to associate an
activity in the grid with a certain quanta of resource such as a
certain percentage of CPU, memory or disk. The other requirement is
recreating the necessary environment (in terms of software
configuration) that the user needs on the grid. Most applications
require a very specific configuration and how it's provided for on a
remote resource in the grid reliably and dynamically is a major
hurdle." In a sense, virtual workspaces take one of the most complex
variables out of the Grid equation, application deployment. Virtual
workspaces will be a key driver in the coming year of Grid adoption.

**SIX INTERESTING ENTERPRISE GRID VENDORS TO WATCH IN 2006**

Platform Computing

Platform wrapped up '05 with the release of the Enterprise Grid
Orchestrator -- a new product for the integration and management of
heterogeneous resources in a Grid environment. 2006 should be a very
interesting year for the Toronto-based company, which partnered with
Microsoft for LSF to offer some key functionality in the Windows
Compute Cluster Server release next year. Songnian Zhou, CEO of
Platform, has pointed out that "The killer app for Grid is not the
components of applications -- ERP, CRM -- but the connection of all of
these components of enterprise applications to form business
processes." Platform is looking beyond the traditional Grid
application set. However, instead of simply looking down another alley
they are thinking in terms of application aggregation, the
commonalities that exist between different application sets and where
this all comes together. This type of thinking is going to be critical
in the coming year.

Univa Corp.

Univa is a rather obvious addition to the 2006 watch list. With the
recent beta announcement of Univa Globus Enterprise (PDF), the company
is officially opening the doors for business and is the first Grid
start-up assuming the services/support/distribution model role for the
open source Globus Toolkit (similar to Red Hat's role supporting
Linux). Not only will Univa satisfy the proverbial "throat to choke"
requirement for enterprise Grid end- users, they also are developing
their own GT features and extensions that will make the Globus Toolkit
truly "enterprise ready." There is also the fact that the founders of
Univa are the definitive pioneers/inventors of the Globus Toolkit.
Steve Tuecke, Ian Foster and Carl Kesselman were also key players in
laying the groundwork for the standards on which most grids are built
-- and bar none, no one "gets" open source Grids (and enterprise
requirements thereof) to the extent that these guys do.

EMC Corp.

Last week's acquisition of Acxiom's Grid software by EMC was perhaps a
harbinger of other big EMC Grid announcements to come in '06. As Grid
technology evolves, the industry is moving away from mere compute
Grids, and toward data Grids -- where data virtualization and storage
challenges could escalate EMC to one of the key outspoken players for
the industry. Pushing on Grid directions is also a way for EMC to stay
competitive with IBM. The rivalry between the two was well-documented
in '05. Network-Attached Storage was one of the key areas in which
these two vendors battled, and for the Grid community, the issue of
coordinated data sharing (including navigating firewalls and pulling
data out of storage devices at the edge) is another frontier area for
the industry that will see a lot of activity in 2006. EMC has had some
interesting personnel additions over the last couple of years that
make it clear they're serious about Grid. Ian Baird (from Platform) is
the company's CTO for Grid and Utility Computing Solutions. Jeff Nick
(former On-Demand heavy hitter from IBM) is the company's CTO. EMC
isn't a company that's just fashionably turning up the dial on Grid
for PR purposes -- they're in it for the long haul. The company's
ownership of VMware -- and the Grid/Virtualization relationship -- is
another reason why EMC will be interesting to keep an eye on in 2006.

Cisco Systems

Cisco truly does put the "intelligence in the network," and the
convergence of virtualization, loosely-coupled services, systems
management and dynamic provisioning capabilities is changing the role
of the network from the mere transport of IP packets to central
nervous system for the IT infrastructure. It would appear that Cisco's
"Intelligence in the Network" mantra will continue to manifest itself
in Grid discussions -- as virtualization concerns continue to move
further down the stack. In '05, we saw Cisco snap up Infiniband
datacenter interconnect TopSpin for $250 million, and with Cisco's
"Application-Oriented Networking" product line and an active "Server
Networking and Virtualization" group, I predict we will see "Grid" and
"Cisco" mentioned in the same sentence with greater frequency in 2006.

Network Appliance

I'm really intrigued to watch the evolution of NetApp's Data OnTap GX
-- the operating system that's driving their virtualization product
evolution. OK, I may have fallen for their marketing hype, hook, line
and sinker, but the fact that these guys are drilling down to the OS
level tells me that they are venturing into uncharted waters with
respect to the complexities of what providing virtualized data really
means. It's more than an application and more than middleware -- it's
a new way of thinking about process control, device management, file
management and networking, all those elements that we learn about in
the first few chapters of any book on operating system concepts.
Another key concept is information management, how files and other
system resources are accessed and controlled. NetApp has mentioned
that another key component of the slimmed down first release of Data
OnTap GX is its unified namespace capability. Clearly, they see that
multiple ways to access the same data and aggregations of data
presented in a unified manner are the keys to virtualization. It's
going to be interesting to see how this is done and if they are
conforming to standards where the rubber meets the road. If the OnTap
GX operating system is effective and becomes a commonly accepted
method of data virtualization, it could prove revolutionary to
enterprise Grid adoption. So, keep an eye on NetApp in '06.

Microsoft Corp.

Bill Gates' keynote at Supercomputing 2005 in Seattle was, at its
root, a "here we are" announcement. Although largely compute
Grid-focused, Gates' talk also highlighted a vision for scientific
workflow and was quick to point out that Microsoft was a driving force
behind XML and Web Services standards. But the importance of Microsoft
is even simpler than that. Microsoft applications such as Excel are
simply accepted standard applications and user interfaces for
computing. A Grid-driven version of Microsoft Excel would indeed be a
killer app. There have long since been many smaller players actively
providing Grid and Grid-like products for Windows operating systems.
It is not hard for one to imagine Microsoft leveraging these companies
in any way its massive bankroll desires. Tony Hey and Fabrizio
Gagliardi joined Microsoft last year. Both are internationally
recognized leaders in Grid computing. Does Microsoft "get" Grid
computing beyond compute grids and cycle scavenging? Will Microsoft
support and embrace the standards that have been so important to the
evolution of Grid computing thus far, or will they simply redefine
Grid as they see fit? Microsoft obviously has something brewing in the
area of Grid computing, and they have a history of proving they can't
be ignored. We'll see if '06 is the year that they open the kimono on
their Grid efforts a little more.

User Object Sharing Lock

Solaris does both spin lock and adaptive lock to reduce the lock contention
In addition, in the SMP box, we can scale the lock by having an array of locks

(1) Merging instead of Locking
(2) spin and adpative locking

Virtualization on Sun Server for Solaris

We will introduce our first generation of virtualization called logical domains
on the T2000 and T1000, to bring people beyond the container technology
Solaris 10 already offers. Running multiple operating systems on a single
computer more efficient.

The logical domain feature will be available in an update to the T1000 and
T2000 firmware, software that runs at a lower level than the operating system.
The update won't degrade performance or require customers to recertify their software.
This new virtualization will ship in August.

Data Pipeline Processing & Data Federation

(1) Data scales from terabyte to petabyte reside in archives
(2) specific dataset accessing, download does not work (too long to download)
(3) Virtual Obseratory --- Grid Interface from Grid Portal
(4) Pipeline processing and data federation
(5) virtual data dynamically created from archieved components
(6) Registry for Metadata of the content for sub data set accessing
(7) samll query againist archieve server, large for cluster of computers
(8) no computation near the data but use cluster of computation near the data
(9) Portal federate web services to provide uniform access

Grid for Predictive Maintenance

Data Centricity

(1) high volume and real time data collection
(2) diverse data repositories

Data Distribution

(1) data monitoring, collection, data storage, mining and fault diagnosis
(2) standard communication and protocols within Grid Paradigm
(3) open data protocols and meta-labling schemes

Grid Enabling searhing

(1) distribute search for scalable, neural-network-derived network
(2) visible on the Grid portal
(3) grid middleware service create the instantiations of the modules
on the set of registered computers
(4) use distributed resources achieve a timely delivery of information

It is a grid based system

AM BasicUser creation template comes up with iplanet-am-user-service

Why AM BasicUser creation template comes with "iplanet-am-user-service" as default OC ?

ou=BasicUser,ou=CreationTemplates,
........
sunkeyvalue=required=objectClass=iplanet-am-user-service
........

System Working Principle

For a server to startup or reboot

(1) bootstrap program needs to run first

bootstrap program is simple and stored in firmware known by ROM or EEPROM
it init all aspect of the system, from CPU registers to devices controllers
to memory contents.

(2) bootstrap program must locate and load OS kernel into memory, in addition
the OS then starts to executing the first process such as "init" and wait
for some event to occur.

(3) occuring of the event is singalled by "interrupt" from HW or SW. HW can trigger
an interrupt at any time by sending a signal to the CPU, usually by the way
of system bus. SW may trigger an interrupt by an operation called "system call"


(4) Interrupt transfer the control to a proper interrupt service routine. When a CPU is interrupted, it stops the current processing and immediately
transfers the execution to a fixed location. The fixed location contains
the starting address where the service routine for the interrupt is located.
The service routine executes, on completion, CPU resume the interrupted
computation.

Server & Shared Memory Access

Server is composed of one or more CPUs and device controllers connected through Bus.
CPUs and Device Controllers are executed co-currently and competing for memory cycle.
Memory Controller is to synchronize accessing to memory so that CPUs and Device
Controllers are accessing Memory Cycle orderly.

AM 7 deployment process over AS8.1 cluster

I have hacked out the manual process for AM7 on AS 8.1 cluster env
I have DAS central controls a cluster of 2 instances of AM7 on another 2 servers
The process to deploy AM instance on any of AS cluster instance is listed
below. Please streamline the processs as development process required.


Please note: All work is done on the AS8.1 instance server instead of DAS hosts

(1) Modify AM web module DD

For amserver web module

in sun-web.xml, replace

DTD Sun Java System Application Server 7.0

as

DTD Sun One Application Server 7.0


Do the same thing for ampassword and amconsole web module



(2) create silent file.

Please note to fill in Server_HOST as AS instance host name instead of
DAS_HOST here


(3) run amconfig -s silenfile

to build DIT and create war files


(4) add permission to AS instance server.policy file

grant{
permission java.security.AllPermission;
};

Please note the permissions which comes
from as81amconfig does not work

(5) manual deploy created web module


asadmin deploy --user admin --passwordfile XXXXXXXX
--host --port
--contextroot amserver --name amserver
--target /opt/SUNWam/amserver.war

The same applied to ampassword etc.


(6) restart the server instance



Thanks

Lei

AM cluster based deployment AMLController.class not found

(1) It seems DD cause the cluster deployment failure
(2) AMLController.class not found. However, I have
found AMLController.class under mobile_services.jar
and .java. class suffix has the mobile_services.jar
in /opt/SUNWma/ dir


but still has AMLController.class not found error

A issues on AS cluster based deployment

(1) It seems DAS based deployment is required
(2) However, of DAS based deployment, it requires all deployment realted
war and dtd files specified in the DOCTYPE of deployment descriptors
are local for DAS which could cause issues, such as AM and PS deployment
(3) Another major issues is that:

if 2.3 version DD requires FQ URI then if the deployment architecture
are located in firewall, http://java.sun.com/dtd/ssssss will not be
fetched for vaildation.

if file protocol based dtd specification within DTD, it requires local
file structure on the DAS system

Cluster based AS deployment issues and suggestions

AS does the 2.3 dtd requirement. web app xml DD specifies file protocol
based DD $install_home/lib/dtds/* and http://java.sun.com/dtd/

It looks for local file system based on file protocol first,
It then looking for http://java.sun.com/dtd/

This works for single host and instance based AS deployment.

What you explain works for the case if single
host and AS instance based deployment. However, for the cluster based deployment
over remote node agent, you can not require (1) DD path exists on DAS host pertaining
to file protocol specified in DD xml (2) FQ URI is accessable within firewall policy.

My suggestion is to cluster based deployment to stream over the DD from node agent
host to DAS in oder to fetch DD content for deployment further processing.

A definitive issue for distributed AM/AS deployment over node agent

I have experienced issues to deploy AM over AS 8.x via remote node
> agent. I have AS DAS admin server and node agent seperated deployment
> use case. AM is deployed from a remote AS instance via node agent
> AS central deployment repository is located at DAS admin server which
> is remote to AM server AS instance.
>
> I am using amconfig -s silentfile post installation configuration. It seems
> AM DIT are able to create in DS repository. However, AM specific web
> app war file has issues
>
> (1)
>
> -- # /opt/SUNWappserver/bin/asadmin deploy --user admin
> --password ll5129427 --host v1280-137-06.mdelabs-mpk.com
> --port 6666 --contextroot amserver --name amserver
> --target jesswitch-cluster /opt/SUNWam/amserver.war
>
> CLI171 Command deploy failed : Deploying application in domain
> failed; Error loading deployment descriptors for amserver
> -- /etc/opt/SUNWam/dtd/sun-web-app_2_3-0.dtd (No such file or directory)
> ; requested operation cannot be completed
>
> Since on DAS server, there is no AM deployment, therefore
> /etc/opt/SUNWam/dtd/sun-web-app_2_3-0.dtd does not exist
> on DAS server. I know I can manually copy over to DAS, any work
> around done to ensure the amconfig works in AS/AM distributed
> environment ?
>
> (2) all other war file deployment shares the same issue
>
> (3) jvm configuration change create-jvm-options
> has the issue too.

D Code stablility & Kernel Module and function calls

I have been practicing DTrace, one of a few things to share


After I have worked on different platforms with different kernel
modules, some times even with grid systems and resource virtualization
such as processors, zone, file systems etc. I have realized that
my D code stability in terms of solaris engineering or portability
and reusability in terms of software engineering is where
I need to spend time on.

(1) different versions of kernel modules such as files systems
and different versions of in kernel function calls

(2) different platform architecture design and implementation such
as cpu kernel structures etc. I felt lucky to deal with VM and
VFS since they do the abstraction for me. However, CPU is
one of the exception

(3) normal host based instrumentation vs virtualization environment
such as zone specific.


This may not only for NFS, but NFS is one of the use cases to I need
to continue to put my thoughts on. However, It may not always
introduce more kernel abstraction and encapsulation and some work
still can be done at D code level so that my D code can be more
extendable to the new platforms, kernel modules and kernel functions
as for the different versions of kernel modules and functions, they
are really the challenges for my D code since it means I need to
keep my D code to sync with the life cycle of your module and
functions releases.

D Code Stability

I have been practicing DTrace, one of a few things to share


After I have worked on different platforms with different kernel
modules, some times even with grid systems and resource virtualization
such as processors, zone, file systems etc. I have realized that
my D code stability in terms of solaris engineering or portability
and reusability in terms of software engineering is where
I need to spend time on.

(1) different versions of kernel modules such as files systems
and different versions of in kernel function calls

(2) different platform architecture design and implementation such
as cpu kernel structures etc. I felt lucky to deal with VM and
VFS since they do the abstraction for me. However, CPU is
one of the exception

(3) normal host based instrumentation vs virtualization environment
such as zone specific.


This may not only for NFS, but NFS is one of the use cases to I need
to continue to put my thoughts on. However, It may not always
introduce more kernel abstraction and encapsulation and some work
still can be done at D code level so that my D code can be more
extendable to the new platforms, kernel modules and kernel functions
as for the different versions of kernel modules and functions, they
are really the challenges for my D code since it means I need to
keep my D code to sync with the life cycle of your module and
functions releases.