HP and Intel Speak at PASS 2005 European Conference

<?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" />

 

The 2005 event is off to a great start.  The attendance stands at about 350 people, including all bodies such as vendors and speakers.  There are attendees here from over forty countries!

 

Day 2 of the event was started with a joint keynote presented by Neil MacDonald of HP and Kevin O’Donovan of Intel, and covered a number of interesting topics about 64-bit computing.  Since 64-bit is the obvious next evolution in computing, this topic was on the top of everyone’s mind.  After all, how many of you were around when we had to do the big migration from 16-bit to 32-bit?  It wasn’t easy!  Here are some notes from Neal and Kevin.

 

Neil MacDonald, PhD (HP)

 

RISC reaching the end of its lifecycle.  Proliants are moving up strongly into the high-end of the market where RISC technology (for example, Integrity Servers, which can scale to 128 processors) usually dominated.  HP is the largest provider of high-end and 64-bit Windows servers.  HP’s high-end servers now post ~786k TPM benchmark at a price/performance rating of $6.49!  That’s more than 100x improvement since 1996! 

 

Memory access is the biggest challenge in high-end servers.  Most high-end servers must use NUMA (non-uniform memory access).  Performance may vary depending on where memory is physically located, such as whether the memory is on-cell or off-cell.  The database will be able to hide that.  Migration from 32-bit is seamless.  This drives down TCO enormously.

 

For example, HP has a large retail customer loading 70m rows (70gb) into a 1tb data warehouse in 12 minutes, with reindexing taking 40 minutes.  On an HP Superdome system, they restored 9tb in 45 minutes, and did table scans at 6.7 gb/sec!

 

Kevin O’Donovan (Intel)

 

Systems are only as strong as their weakest part.  Core processor is very important, but Intel also pays a lot of attention to those subsystems that surround the core CPU such as I/O, memory, and power management.

 

The new Intel multicore chips have literally two CPUs on a single silicon wafer.  It allows you to parallelize processes that were never possible in the past.  Xeon (32-bit) has extensions to virtually address more memory.  Itanium (64-bit) is fully configured for explicitly parallelized instruction computing (EPIC) processing.  Testing has shown big boosts (+30% on TPC, +40% on <?xml:namespace prefix = st1 ns = "urn:schemas-microsoft-com:office:smarttags" />SAPSD, +44% on SPECint_rate).  Plus has much greater reliability and scalability (+59% on 8 way, +90% on 32 way, many more on 64 way).

 

Montecito, the next version of Itanium, is due later this year and carries around a ~70% performance boost, but also uses less ~30% power and generates ~30% less heat than the current generation of Itanium.  (If you’ve ever stood near a rack of servers in a data center, you know how hot these things can get!)  Itanium releases further in the future include Montvale (2006) and Tukwila (2007).  In 2007, a 4-CPU server will be able to replace a current 64-CPU server!

 

The good news is the Microsoft has paid the ‘tax’, if you will, on new applications.  So you can start to work immediately on 64-bit using SQL Server without need to recompile.

 

Customers can benchmark their custom solutions at HP’s customer testing lab in Sophia Antipolis in France and Boeblingen in their benchmarking and demo center in Germany (http://www.hp.com/benchmarkcenter).  The Boeblingen center also has competency centers for all major hardware areas including storage, CPUs, and so forth.  Intel is developing new technologies to support auxiliary features of the processor.  For example, the new high-end chips enable much greater use of server virtualization.  This necessitates the development of new security features such as disabling any virtual server on the server from snooping the data used by another virtual server on the same box. 

 

Intel has three solution centers across Europe as well.  Locations are Munich, London, and also in Stockholm, Sweden.  You can “try-and-buy” hardware that you might be considered for you enterprise applications.  This is especially important when doing proof-of-concepts for high-end enterprise applications.  These centers offer the greatest value of any testing opportunity because you’re testing your own application and not a standardized benchmark.

 

Neil - www.hp.com/products1/itanium/testimonials/index.html describes many case studies for how large enterprises are using the new technology.  Customers are doing things like data warehousing, ERP, and database consolidation.  Raymond James Financial, for example, achieved a 5x increase in performance.

 

Whew!  Now my fingers are tired.

 

I have one more session today – ‘Bare Metal Tuning’ and then it’s time to pack and head home.

 

I certainly have enjoyed my time in beautiful Munich and hope to come back some day!

 

-Kevin