Edit: Some links no longer work
Originally posted August 6, 2019 on AIXchange
In this AIXchange blog, Technical Editor Rob McNelly recaps a presentation by IBMer Nigel Griffiths.
Nigel Griffiths recently gave a presentation on the long-term viability of IBM Power Systems hardware and the POWER processor. He then summarized his points here:
- AIX is the UNIX winner in Technology and Performance—IBM having out Researched and out Developed all the other UNIX servers from other vendors over this time.
- Linux on POWER performance lead—POWER9 2+ times faster than Intel per core
- Accelerated AI and HPC Workloads on the AC922 give world class and the current number 1 and number 2 in the Top500 HPC computers List
This is only a small part of his post, but obviously the idea that IBM servers represent an end-to-end solution is compelling.
Of course, processors don’t make a server. With Power Systems IBM makes the servers too and we make sure new POWER processors are designed to make powerful servers by enhancing everything else too: memory, disks, adapters and all the interconnecting buses.
Finally, he lays out the investment IBM—and just a couple of its largest customers—have made in this technology:
Is IBM going to ditch POWER?
- Power Systems is a large successful business. I’m not allowed to detail the numbers of specific parts of IBM (some details are in the IBM Annual Statement) but shall we say it is in the many billions of dollars scale. IBM is not going to shut that down on a whim.
- [About] four years ago, I was asked by a worldwide banking client to audit a particular workload on POWER7+ as it was about to move to POWER8. It was only ~ 25 CPU cores but lots of RAM and was running nicely but pretty busy.
- I was to predict how it would run on POWER8 and any recommended tuning changes.
- At lunchtime, the client visited my cubical to explain this was an important workload.
- If the workload server failed for an hour, it would seriously impact the bank’s reputation.
- If it was not available for a whole day the world economy might not recover!!!
- “No pressure,” she said as she walks back down the corridor laughing loudly!
- I started triple checking all my figures!
- This is the sort of workload clients insist on running POWER servers—it has to fast and it has to be reliable.
- The server was actually running in a quadruple backup server arrangement and across countries.
3. I am now working on a soak test for a High-Performance Cluster at a racing team (can’t say more). The soak test involves taking the server to 100% busy for 10 days with a 2 day cooling down period in the middle. This, I think, is due to unreliable previous hardware. The client is moving to POWER9 for high performance and super high RAS = Reliability, Availability, Serviceability. It does not breakdown; if in the unlikely case of a break down it stay up running (disabling the fault components); then the faulty components can be replaced online or at a later convenient time). With Live Partition Mobility (LPM) running workloads can be moved while running to a different server. This is all normal for Power Systems.
While I’ve previously explained why AIX isn’t going away, I want to make you aware of Nigel’s thoughts and some of the reasoning behind them. He covers much more ground, so be sure to read the whole post.