At IDC’s annual ISC breakfast there was a good deal more than market update numbers although there were plenty of those: “We try to track every server sold, every quarter, worldwide,” said Earl Joseph, IDC program vice president and executive director HPC User Forum. Perhaps more revealing and as important this year was IDC’s unveiling of significant changes in how it is characterizing and measuring the HPC world.
For example, GPU and accelerator tracking has been promoted to a formal activity. But first the top line: The HPC market is growing after a period of sluggishness. It’s now around $10 billion (servers) and IDC expects it to grow substantially this year and steadily through 2019. Also, the collision of HPC and big data has perhaps been bigger than expected and prompted a change in IDC’s data gathering efforts.
“Lately we found that the vendors are having great difficulty determining how many systems go into HPC versus other areas so we’ve had to switch our data collection to very large scale surveys of buyers and users. Last year we exceeded 40,000 surveys and this year it will probably be on the same order of magnitude,” said Joseph.
“One thing I want to mention is about six months ago, with all the data collection we were doing and comparing things, some numbers just stopped adding up and what we found was that the financial sector has been growing faster than we thought and has become much larger over the last two years. In 2010 it didn’t recover much but then went into a hyper growth mode so you will see us restate our numbers for the last two years on the financial sector [up] on the order of at least a 50 percent,” the analyst continued.
For years, IDC has divvied up HPC into four categories: supercomputer; divisional; departmental; and workgroup. Those numbers are shown below.
However sales for the top ten systems skew cumulative data so much that IDC is changing the way it characterizes the market.
“Except for the top ten systems the whole market grew dramatically. This is something we’ve struggled with in the past couple of years. The plan now is to introduce a fifth competitive segment all round the top ten. We’re not sure what we are going to call it, maybe something like leadership computers, but the market dynamics around those top 5-10 machines are totally different,” said Joseph. The report card and forecast the broader market are shown below.
You can see that storage remains hot and is expected to grow faster than other segments; that is being fueled by more data collection and more data analysis. IDC has labeled the fragmented storage market as something of a Wild West, although big players are now turning their attention to that market. IDC suggests similar dynamics in the interconnect space. Middleware is also being watched carefully not least because one would expect increased tool buying to accompany any large-scale movement to upgrade software; the latter of course is seen as a growing pain point in HPC.
Well-noted was the much discussed effect of IBM’s sale of its x86 business to Lenovo. The table below on Q1 2015 sales starkly portrays IBM’s tumble. Joseph commented: “The industry for the last decade was basically tied with two vendors: HP and IBM had 30-34 percent each. Then you had Dell at half of that. All the other vendors had a couple percent or lower. The market now is completely different. HP is the market leader; it’s on the order of a third of the market. We were expecting to see three vendors around 15 percent and the shock here is the IBM decline was higher than expected.”
Joseph was quick to add caution in reading Cray, SGI and the smaller vendors’ one-quarter numbers: “Cray’s going to have a fourth quarter that is three times their market share and so you really have to average the small vendors.” 2015’s Q1 showed ten percent growth relative to 2014 Q1 (See, IDC Says Lenovo Strengthens, IBM Stumbles in Q1 Ranking).
To a fair degree, ongoing market dynamics and IDC’s responses to them were as interesting as the sales numbers. Big data continues to expand and transform the HPC world; cloud adoption for HPC purposes is probably higher than you thought; nascent HPC ROI models are showing dramatic value; and vendor shuffling, perhaps not unexpected following IBM’s x86 business sale, continues. A plus for IBM is emerging traction in the market for OpenPOWER which now has nearly 150 members; over time IBM expects to regain market share in the technical server market along with other members of OpenPOWER.
Among the top HPC trends and watch areas cited by IDC are:
- 2014 was soft but 2015 won’t be.
- Big data combined with HPC is creating new solutions.
- Software issues continue to grow.
- GPUs & accelerators extend their impact.
- Non-x86 processors could alter the landscape.
- China looms large(r).
- Growing influence of the datacenter in IT food chain.
- HPC in the cloud gaining traction.
The cloud numbers are interesting. IDC surveyed 157 HPC sites on their use of clouds and roughly 25 percent reported they were using clouds and those who use clouds reported that nearly a third (31.2 percent) of their workloads were run on clouds.
Bob Sorensen, research vice president in IDC’s High Performance Computing group, said, “I should add the caution that some of what they are uploading are what we would consider the low hanging fruit, some of the embarrassingly parallel applications that don’t require very specific architecture. So it’s still early but we expect more sophisticated applications [to be run in the cloud] as time progresses.”
Perhaps the most transformative trend explored is big data convergence with HPC. Just what that means varies widely depending on which vendor is talking, influenced in no small measure but the type of technology they have to sell.
Interestingly, IDC is changing how it examines HPDA activities and introduced four new industry/application workflows including:
- Fraud and anomaly detection. This “horizontal” workload segment centers around identifying harmful or potentially harmful patterns and causes using graph analysis, semantic analysis, or other high performance analytics techniques.
- Marketing. This segment covers the use of HPDA to promote products or services, typically using complex algorithms to discern potential customers’ demographics, buying preferences and habits.
- Business intelligence. The workload segment uses HPDA to identify opportunities to advance the market position and competitiveness of businesses, by better understanding themselves, their competitors, and the evolving dynamics of the markets they participate in.
- Other Commercial HPDA. This catchall segment includes all commercial HPDA workloads other than the three just described. Over time, IDC expects some of these workloads to become significant enough to split out, i.e. the use of HPDA to manage large IT infrastructures, and Internet-of-Things (IoT) infrastructures.
(Note: financial, classified buyers, etc. will continue to be listed under the existing IDC segments)
Notably, “HPDA adoption has changed over time. It’s made its way into the scientific community and basically reached the point where it stands almost shoulder to shoulder with traditional modeling and simulation,” said Sorensen. Right now there are few rules, and IDC’s advice is to ‘embrace the chaos.’ There is so far no single best or even standard HPDA solution.
Joseph added that in the last six months IDC has conducted seven HPDA surveys, “We’ve studied everything from what are the underlying applications, what are the algorithms, where are the benchmarks, how do you evaluate a good HPDA system and solution, whether a distributed database or single database is better, and what hardware architecture do vendors plan to use to address the HPDA space.”
Lots of questions and opportunities think IDC and virtually everyone else. In a survey of 128 HPC sites about 23 percent reported use of an HPC system for HPDA purposes.
One intriguing topic tackled is HPC ROI measurement. IDC has been developing economic models to try to unravel that question for some time, mostly driven by a contract with DOE. That three-year effort will be likely extended another 6-9 years, said Joseph.
In simple terms, the model input is HPC investment and the output is revenue growth, profit, and job creation. The early work seems exceedingly positive. IDC reported that for every $1 invested in HPC $356 in revenue and $38 in profit were generated – impressive if accurate.
Many more topics were touched upon and IDC is making the report available to all of the session attendees. This was the first year the event was part of ISC’s official agenda. Joseph reviewed HPC User Forum activities and said the group now planned to make all six year’s worth of HPC User Forum meeting presentations available to everyone, free, from the HPC User Forum website. It would be a fascinating exercise to review HPC expectation (user and vendor) versus results.