IBM/Intel Smash Processing Records

[butitoy]

---

Enterprise computing takes a big step forward.

In the world of enterprise computing, systems must exhibit three characteristics: speed, availability, and scalability. Failure to satisfy any of these three will result in costly inefficiency at best and, at worst, outright catastrophe. This is especially true in the banking industry, where systems must be able to handle many millions of transactions daily, essentially never fail, and be able to grow in capacity as the enterprise expands its services and client base.


The ability to hit all three of these requirements took a big step forward recently when IBM and Intel announced test results showing SAP Transaction Banking (TRBK) daytime processing of 56,518,000 account postings per hour. Night processing results hit 22,382,000 balanced accounts per hour. This is a 3X leap over the prior world record for these tests.
Speed advance? Check.
These results were achieved on an IBM platform of IBM System x3690 X5 servers running IBM DB2 pureScale clustering. The dual-processor servers all ran Intel Xeon E7-2870 (2.4 GHz) CPUs. With 10 cores per chip, that yielded a total of 20 cores (and 40 threads) per server. IBM used five servers, each with 256 GB of memory and 6 Gb/s SAS drives running on a QDR InfiniBand storage fabric. To simulate all of those users, IBM clustered 21 HS22 application servers in two Blade H chassis.


The clustering is where availability and scalability come in. IBM’s second breakthrough was bringing its DB2 pureScale technology to its BladeServer architecture for the first time. Blades represent a far greater value proposition for data centers owing to their higher compute density and power efficiency. Blades lend themselves far more easily than conventional rack servers to redundancy and hot swapping in the event of system failure or upgrading.
Additionally, a blade chassis makes scaling a snap. IT might start with a 14-bay enclosure, such as IBM’s BladeCenter H, but only start with six or seven (or 11, in this case) blades. That leaves plenty of open bay slots and room to add processing capacity as needed. With only a 9U rack height, this blade enclosure can not only save costs compared to conventional 2U or 4U servers but also reduce overall compute footprint in the data center.


The fact that such economies can be realized as part of a record-breaking test bed only adds to the story. While IBM and Intel have succeeded in shattering prior performance records in a key enterprise sector, perhaps the greater take-away is that these performance advances can now be had in more economical clusters that will help businesses conserve their resources.


http://www.tomsitpro.com/articles/blade-servers-performance-records-transactions,1-27.html

[butitoy]

powered by 10core Xeon | 20 threads.

an experience on performance that an istoryan, Sir Simoko had posted " Orgasmic! " when asked how is it.

( taken from an old thread here )

[SantosJB]

wow! galing naman.

[istoryanimo]

fits for company like google, facebook, yahoo, ug uban pa.