by June 30, 2004 0 comments

The highest-end servers from IBM are based on its POWER architecture. It’s being used for building the highest-end super computers as well as high-end servers for commercial deployments. For instance, IBM is designing the fastest high performance server of the world, called ASCI-purple, which will cross the 100 Teraflop threshold. Two of these will be delivered to the US department of energy and will have more power than all the computers combined in the top 500 list of High Performance Computers. The commercial counterpart based on the POWER4 architecture (one of the latest generation) is the pSeries 690 server. ?

The pSeries uses Chipkill memory and FFDC techniques for better reliability

The system is expandable upto 32-way SMP using 1.5/1.7/1.9 GHz POWER4+ processors, upto 512 MB of L3 cache, and 1 Terabyte of RAM. It can house up to 128 hot-swappable disk bays using 7 additional drawers taking it upto 18.7 Terabyte of storage per node. It can also take upto 160 PCI/ PCI-X adapters.

The system not only provides raw power, but also some advanced reliability features that are worth a mention since they relieve the system administrator of his worst nightmare, downtime.

Chipkill memory – An advanced form of ECC memory technology, it protects the systems from any single memory chip failures as well as any number of multi-bit errors in the memory. The ECC technology on the other hand can only correct single-bit errors and detect multi-bit errors.

First Failure Data Capture (FFDC) – It is a technique used for collecting data that will help in diagnosing the problem without reproducing it. Some of the information that FFDC captures can be trace logs, message logs and dumps of in-memory data structures.

Dynamic Processor De-allocation – A system report is generated by AIX when the number of processor errors hit a threshold and it recognizes that this processor will fail, the firmware generates an error report. In a multi-processor system AIX stops using this untrustworthy processor and de-allocates it from further usage. ?

IBM p690, p680 and p620 AT United Bulgarian Bank

Formed in 1992 with the merger of 22 local banks, UBB was privatized in 1997. It serves more than 1 million personal and about a 100,000 business customers. It operates almost 50% of Bulgaria’s ATMs, 37% of all debit cards, and 19% of money market trading.

The challenge in this case was to integrate the disparate systems that were operated by all these merged banks. These systems were replaced by IBM eserver pSeries p690, p680 and p620. The main production server is a single p690 divided into three logical partitions (LPARs), with p680 and p620 servers running supporting systems and databases. Each production partit\ion on the p690 is designed to failover to a separate p680 or p620 server.The p690 contains 26 processors out of which 10 are assigned to one LPAR running Temenos, a leading banking solution, and the other two smaller partitions of a single processor each are dedicated for card services and SWIFT payments. A single system has helped the bank in streamlining its operations and has reduced the number of system administrators, looking after servers with a much smaller physical footprint.

No Comments so far

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.