This document lists the new features that are available in this release of Windows® HPC Server 2008 R2.
If you want to test some of the new features listed in this document, see the Test Focus Guide for Windows HPC Server 2008 R2 Beta 1, and the Developer Resources for Windows HPC Server 2008 R2 Beta 1.
Job scheduling and runtime
The following job scheduling features are new in this release:
- Service-balanced scheduling. This
release includes a new scheduling mode that optimizes the process
of starting jobs, and balances in real time the resources that are
assigned to jobs, according to their priority. To enable this
functionality, run the following command on the head node of your
cluscfg setparams schedulingmode=balanced.
- Dynamic parametric tasks. This feature
improves the performance of large parametric task sweeps by
creating tasks when they are needed, instead of creating them when
the job is submitted.
- Job progress viewing and reporting. A
redesigned user interface on HPC Cluster Manager displays job
progress. Also, changes to the application programming interface
(API) help you to report more detailed job progress information for
your HPC applications.
- Improved job and task troubleshooting.
The user interface for job and task troubleshooting has been
improved to make it easier to review errors. To see detailed
information about a failed job, double-click that job on HPC
Cluster Manager or HPC Job Manager.
- Task cancellation grace period. In
this release, when you cancel a task that is running, the HPC Node
Manager Service stops the task by sending a CTRL+BREAK event
to the application. The application then has a preset period of
time to exit gracefully. The application can use this time to save
state information, write a log message, create or delete files, or
finish the current service call. You can configure the task
cancellation grace period by changing the
TaskCancelGracePeriod cluster property.
- New node preparation and node release
tasks for jobs. These new job tasks are available in HPC
Cluster Manager, the API, HPC PowerShell™, and the command-line
tools. A node preparation task runs on a node before any other
tasks, and you can use it to creating local folders, copying files
onto the node, or performing other initialization steps. A node
release task runs on a node when the node is ready to be released
from the job, and you can use it to delete temporary files, copy
local files to a central shared folder, or perform other cleanup
- Scheduling at scale and better task
throughput at scale. This release includes support for larger
clusters, more jobs, and larger jobs—including improved scheduling
and task throughput at scale.
- Update user credentials once for all
queued jobs. This release includes improvements to the way that
user credentials are handled. You can now update your user
credentials once for all your jobs that are in the queue.
SOA scheduling and runtime
The following runtime features are new in this release:
- Support for multiple clients in SOA
sessions. In the first CTP release of
HPC Pack 2008 R2 (CTP 1), a service-oriented
architecture (SOA) session could only be used by one client
application at a time. Multiple client applications sharing the
same session could not determine which results belonged to them. In
this release of HPC Pack 2008 R2, by improving the
session API, every batch of computation is tagged with a GUID so
that multiple client applications can share the same SOA session by
identifying its own GUID when sending computation requests and
- Fire and recollect programming model.
This release supports offline SOA applications, including client
disconnect (SOA batch) and client resilience.
- Single-job SOA sessions. In this
release, it is no longer necessary to have two jobs running for
each SOA session (one job for the SOA session broker, and one for
the service). This one-to-one mapping between jobs and sessions can
make monitoring and reporting simpler.
- Built-in flow control. The new SOA
session API and the new broker Web service interface in this
release include a built-in flow control. With this feature, you do
not need to implement your own throttling behavior, as was
necessary in previous releases.
- Better Java interoperability. This
release provides improved interoperability with Java client
The following deployment features are new in this release:
- Deployment at scale. This release
provides efficient deployment of up to 1,000 nodes.
- Remote database configuration. During
the installation process of HPC Pack 2008 R2, you
can configure Microsoft® SQL Server™ 2008 SP1 databases
on computers that are not the head node of your HPC cluster, and
then use them to host cluster management, job scheduling,
reporting, and diagnostics information.
- iSCSI Deployment. With this release,
you can deploy nodes on your cluster that boot over the network by
using an iSCSI connection. This new feature helps you to centralize
your storage, and to deploy diskless nodes (that is, computers that
do not run the operating system from a local hard disk drive or
that do not have a hard disk drive installed).
The following cluster management features are new in this release:
- Location-based node management. With
this release, you can quickly view compute nodes in HPC Cluster
Manager based on their location information. To see the new list
view: in Node Management, under Nodes, click By
Location. You can also view compute nodes on the heat map
grouped by their location information. For more efficient viewing
options, you can specify up to three levels of location detail for
each compute node: primary, secondary, and tertiary.
- Heat map enhancements for large
clusters. This release includes improvements to the heat map
view in HPC Cluster Management, to support the display of large
clusters. You can see the status of up to 1,000 nodes without
scrolling. Additionally, you can configure color-coded overlays of
information in the heat map view, and have a prioritized metric
- Tabbed views. With this release you
can create multiple customizable tabs in HPC Cluster Manager. This
enables you to have several list and heat map views available at
the same time, each one displaying the metrics that you select.
- Diagnostics extensibility. You can
create and add your own custom diagnostic tests to
HPC Pack 2008 R2. These tests are run by the HPC Job
Scheduler Service, and you can create and manage them by using HPC
Cluster Manager and HPC PowerShell. You can also specify parameters
- Reporting extensibility. This release
includes a rich reporting database and the ability to create and
add custom reports to HPC Pack 2008 R2.