IBM United States
Software Announcement 202-084
April 8, 2002
IBM Cluster 1600 Now Includes IBM pSeries 690/670 Offerings
(Corrected on May 6, 2002)
Changed and added the Software Requirements for Clusters with pSeries
At a Glance
The Cluster 1600 system now offers:
IBM Cluster 1600 can now be configured with the pSeries 690 and pSeries 670 in addition to the already supported pSeries 660 models, pSeries 680, and RS/6000® SP, in switched and non-switched environments. Switched environments include the SP Switch or the SP Switch2. The SP Switch2 environments include single- and two-plane configurations.
The Cluster 1600 software expands the list of hardware building blocks to include support for the pSeries 690 and pSeries 670 including:
For all servers with machine type 7026, including the pSeries 660 Models 6H1/6H0 with the new 750 MHz feature, scaling is increased to 64 servers.
The RS/6000 SP Switch Frame family is expanded to include new models that can be mounted in IBM's 19-inch racks, enhancing the available switch building blocks of the Cluster 1600.
The following licensed programs are the software building blocks that deliver support for the pSeries 690 and pSeries 670 server:
Note: Refer to the Software Requirements section for the service levels needed to provide announced support.
For the Cluster 1600 Type 9078, one or more of the following building blocks subject to the scalability limits announced:
Planned Availability Dates
April 26, 2002
The IBM Cluster 1600 gives you a single cluster serial number when ordering machine type 9078 Model 160, which simplifies cluster administration and makes it easier for you to track and manage your cluster over time. There is no cost for ordering the feature for the single serial number 9078-160.
The Cluster 1600 is enhanced to include more pSeries 660 models and now, the pSeries 690 and the new pSeries 670. The pSeries 690 is a high-end, 32-way, Giga processor POWER4 system. The new pSeries 670 brings advanced POWER4 technology to the mid-range server environment. The physical components of the pSeries 690 and pSeries 670 can be divided into up to 16 LPARs within one physical server. Each LPAR functions as an individual node within that server. These servers require AIX 5L V5.1, or later, and PSSP V3.4.
The pSeries 690 and pSeries 670 are supported as a Cluster 1600 building
block with the SP Switch2, the SP Switch, or no switch. This support
requires a fix level update on PSSP. A Cluster 1600 system with PSSP is
limited to the 32-bit AIX kernel only. This limits pSeries 690
LPARs to 96 GB of memory.
Scaling Statement for Clusters
A Cluster 1600 with PSSP can consist of 2 to 128 AIX operating system images. An operating system image, or logical node , can be:
Logical nodes are limited in a cluster running PSSP as described in the
7040 Cluster Limits table and the scaling rules that follow. Any cluster
system that exceeds any of the announced supported logical node limits
requires a special bid.
7040 Cluster Limits
SP Switch2 (total switched plus non- Maximum SP SP switched, subject Switch- Values Switch Switch2 to other limits)(1) less pSeries 690/ 2 16 16 16 pSeries 670 servers per Cluster LPARs per 8 8 16 16 pSeries 690 server LPARs per 4 4 16 16 pSeries 670 server LPARs 6 48 48 48 per Cluster
Additional Information SP Switch2 (total switched plus non- Maximum SP SP switched, subject Switch- Values Switch Switch2 to other limits)(1) less Number of 1 1 or 2 1 or 2 0 switch planes supported Number of 2 4 4 4 pSeries 690/ pSeries 670 servers per HMC(2)
Complete configuration restrictions will be documented in the PSSP and
update and will include specific node and server combinations/limits.
Any cluster systems that exceed the announced supported server limit must
be special bid. For additional pSeries 690 scaling, refer to the
Statement of Direction
HACMP 4.4.1 ES and ESCRM features now support LPARs on a pSeries 690 and
pSeries 670. HACMP 4.4.1 ES and ESCRM features now also support the AIX
5L 64-bit kernel on 64-bit capable hardware models, including the pSeries
690 and pSeries 670. A 64-bit kernel is not supported by HACMP 4.4.1 HAS
and CRM features nor for any HACMP running on a PSSP cluster.
The following service is needed for HACMP 64-bit support:
The following service is needed for HACMP LPAR support:
The following service is needed in support of clusters with pSeries 690/670 servers:
HACMP support of the 64-bit kernel, when run on any system with PSSP, is not supported since PSSP supports only the 32-bit kernel.
The establishment of a trusted network between the control workstation (CWS) and the Hardware Management Console (HMC) is needed to bypass a security vulnerability in Cluster 1600 systems with pSeries 690 or 670 servers. Refer to the PSSP V3.4 Read This First document at the Web site:
Accessibility by People with Disabilities
The following feature supports use by people with disabilities:
e-business Infrastructure and Workload Consolidation
For large corporations (over 1000 employees) focused on lowering IT
acquisition and operating costs, the Cluster 1600 helps them build,
run, maintain, and continuously adapt hardware, software, and network
infrastructure to support a broad range of e-business applications common
in the enterprise. With the addition of pSeries 690/pSeries 670 as
Cluster 1600 building blocks, IBM's proven cluster system management
software, PSSP, can now manage copies of the AIX operating system running
on either LPARs of a pSeries 690/pSeries 670 or on a full system
partition. LPARs give the customer the flexibility to change the
configurations easily to adapt to changing workload patterns and capacity
requirements. That, coupled with the current SP environment, gives you
extreme flexibility and investment protection. Customers can assign
workloads to different LPARs of a pSeries 690/pSeries 670 or to other SMP
servers in their Cluster 1600, thus simplifying management of their IT
infrastructure and reducing total cost of ownership.
High Performance Computing
In High Performance Computing, the POWER4-based pSeries 690 offers
outstanding floating-point performance and extreme SMP scalability. The
world's largest, most challenging scientific and technical applications,
like weather forecasting or other large-scale simulations, need vast
amounts of computational power. A single pSeries 690 server (1.3 GHz) is
designed to deliver peak performance of 166 gigaflops. For massively
parallel, message-passing workloads, a cluster of 16 pSeries 690 servers
can deliver peak performance of 2.6 teraflops. With IBM's highly
scalable, highly reliable cluster file system, GPFS, a cluster of pSeries
690 servers can share vast amounts of data among different engineering
workgroups in a large automotive or aerospace engineering firm. As such,
job throughput can be maximized and job turn-around minimized.
Clustering the innovative and powerful pSeries 690 is the latest step in
the long tradition of IBM excellence in high-performance computing.
Large Scale Data Warehousing
Increasingly, large corporations are using information technology to improve customer focus. The ability to collect and efficiently analyze increasingly large amounts of operational and third-party market data is critical to improving product and service offerings, customer satisfaction, and ultimately competitive advantage. Following in the tradition of the RS/6000 SP, the Cluster 1600, configured with either pSeries 690 or pSeries 660-6M1 servers, offers an extremely scalable, highly reliable platform for hosting extremely large and growing corporate data warehouses.
Statement of Direction
Increased pSeries 690/670 Scaling
Support for clusters with as many as 32 pSeries 690/670 servers is planned to be provided by third quarter 2002.
IBM intends to provide support for a maximum of 128 LPARs in a cluster with a maximum of 128 instances of the operating system and a maximum of 32 pSeries 690/pSeries 670 servers.
All statements regarding IBM's future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only.