POWER to the People
Today’s blog is courtesy of Mark Olson, WW Product Offering Manager for Power Systems, who teams with IBM developers to prioritize, announce and roll out new generations of Power Systems hardware. His post complements the latest chapter to be released on facebook in celebrating 25 years of IBM i. Enjoy it, and then join the conversation.
I’m honored to have been asked to contribute a few words about some of the amazing hardware IBM i clients can use today. Power System hardware offers capabilities, capacities and price-performance levels that were almost unimaginable 25 years ago.
The core (pun intended) is the POWER chip. It’s interesting to note that, unlike Watson, which used off-the-shelf POWER7 hardware, achievements like Deep Blue used custom designed/built servers. Deep Blue defeated the champion chess master, Garry Kasparov, in 1997 in a six-game match using custom chips that were specifically built to analyze chess moves. You wouldn’t use Deep Blue to run your business. POWER chips, on the other hand, were built to handle an immensely wide range of uses.
The POWER chip technology running IBM i was the output of many, many developers–including teams of IBMers from Rochester and Austin. Prior to POWER, both the AS/400 and the RS/6000 servers had their own chips that were similar, but different. Development teams collaborated over several years to incrementally evolve the chips and underlying operating systems to accomplish a blended, optimized chip that could run IBM i, AIX and Linux operating systems.
It was important to ensure the key values of both the AS/400 and RS/6000 were implemented. One of the key attributes implemented specifically for IBM i and its single-level storage was address protection. This is sometimes called the “65th bit”. For you technology purists, this isn’t really an additional bit in the chip, but is a tagged pointer implemented in the chip and its execution units. It means that the hardware prevents addresses from being compromised. It’s a key reason why the underlying security and protection of IBM i is so good.
Today’s POWER7+ chip is a marvel of engineering and manufacturing. Thirty-two nanometer technology. More than 2 BILLION transistors per chip. L2 and L3 cache on the processor chip. Up to 8 cores per chip. Ranging from 3 to 4.4 GHz. Wow.
Power Systems servers are of course more than just the POWER chip. Every generation of chip is supported by an integrated set of related hardware. This includes plenty of fast memory, memory controllers, memory buses, I/O buses, PCI slots, power supplies, service processors, expansion drawers, suites of PCI cards, an assortment of HDD and SSD, etc, etc, etc. And, of course, it depends on great architectures tying them together.
Let me highlight a couple of really sweet capabilities. IBM publishes a GHz value associated with each server. For example, a Power 720 can be 3.6 GHz. But the chip doesn’t always run at 3.6 GHz. If you are running extremely high workloads on the server, it will automatically take its temperature and, if cool enough, boost the GHz a few percent higher to help handle the peak workload. Or conversely, if the server is not busy, several options are available to the system. If some of the processor cores aren’t being used at that point in time, they may slow down or even take a nap for a few milliseconds. This is done on a per-core basis, not a per-chip basis. The system does this without impacting your performance. If you authorize the system to do so, work that doesn’t have to be done as quickly as possible will be handled with lower GHz. For example, if you have a batch job which is due at 6 a.m., but it’s currently 1 a.m. and only needs two hours to complete, you can tell your server it’s OK to take longer and it will run at the lower GHz, saving energy and cooling, and still finish before 6 a.m. That’s a “cool” capability.
Another new technology that’s becoming more and more popular is flash memory, also known as solid state drives (SSDs). This is rapidly evolving and its performance, reliability and price/performance, already very good, keeps getting better and better. Today you can configure flash memory on Power Systems servers in a variety of ways. It can be in an “external” storage device like a SAN attached through a Fibre Channel connection, or it can be an “internal” storage configuration leveraging IBM’s premier SAS infrastructure. Three different “internal” configuration options are available to match different price points and other selection criteria. All of the flash memory options I’ve described can make a huge performance difference if your server has an I/O bottle neck or if you want to implement a new application for which I/O performance is critical. The nice thing about IBM i and SSD is that IBM i has the highest level of integrated capability of any operating system to effectively use SSD. For example, in addition to a user being able to assign data base objects to SSD, IBM i can determine what’s hot and what’s cold and, while the user application is running, migrate the hot data to SSD and the cold data to HDD. IBM i can even do that if the SSD is in a SAN. Or IBM i can leverage the Easy Tier capability of IBM SANs to implement hot/cold data across SSD/HDD. You have a lot of flexibility.
I could go on, but I won’t. Thanks for letting me share a few thoughts. Happy Birthday IBM i!!!