Saturday, 26 November 2016

Secrets, lies, women and money: the definitive summary of SC16 - Part 1

Just over a week ago 11,000 people were making their way home from the biggest supercomputing event of the year – SC16 in Salt Lake City. With so much going on at SC, even those who were there in person likely still missed a huge proportion of what happened. It’s simply too busy to keep up with all the news during the week, too many events/talks/meetings happening in parallel, and much of the interesting stuff only gets talked about behind closed doors or through informal networking.

There were even a couple of top-notch tutorials on HPC acquisition and TCO/funding models :-)

Amongst this productive chaos, I was flattered to be told several times at during SC that people find my blogs worth reading and commented they hadn’t seen any recently. I guess the subtext was “it’s about time I wrote some more”. So, I’ll make an effort to blog more often again. Starting with my thoughts on SC16 itself.

As ever, while I do soften the occasional punch in my writing (not usually in person though), there remains the possibility that some readers won’t like some of my opinions, and there’s always the risk of me straying into controversy in places.

I've got four topics to cover: secrets, lies, women and money.

Secrets


It came out in several conversations during the week. “What has been your highlight so far?” “What’s the most grabbing announcement you’ve heard this week?” “What’s the big story of the show?”.

Sadly, my answer was mostly “meh – nothing really”. Many people I spoke to agreed, SC16 was lacking in “big” news. There was lots of “news” but there wasn’t much beyond business as usual, evolutionary announcements, or re-announcing of products that had already been talked about for a long time, etc. There were some desperate attempts by marketing departments to say otherwise (see “Lies” below) but overall SC16 was a calm news week. (My media friends have probably just hit the screen at the word “calm” – I hear they get a bit flooded in the run up to each SC/ISC.)

However, away from the public news, in the off-the-record chats that happen as friends bump into each other when moving around the SC16 campus, and in the many networking receptions, and in private meetings held under NDA, there was plenty of news. Re-read the previous sentence to see why I’m not going to reveal most of these snippets. But I will hint at some.



On the topic of exascale systems, I heard three interesting rumors (well, maybe stronger than mere rumors). Two of these concern the US exascale efforts, and both are likely to please some folk but infuriate others. The third is not likely to shock anyone “in the know” but counters the common public perception of the exascale effort in question. I think the “consensus” prediction for the first exaflops system is that it will be after 2020 and will be in China. As many of you will know, I believe at least part of that consensus will turn out to be wrong and what I heard this week adds weight to my suspicion.

There was some news about a technology vendor hoped by many to make a substantial impact on choice in the market in the coming years. I'm sure the news will break publicly soon, but it hasn't yet as far as I know. Whilst many who heard the scarce details were drawing conclusions of doom, a smaller number who professed to have been more fully briefed predicted a much more positive outcome. Elsewhere, some private conversations suggested another candidate for delivering greater choice to the HPC community might (only might) bring off a powerful development in the next year or so. Overall, hopes for greater choice in the HPC market are looking better.

Away from the technology, there were some interesting heads-ups on people moving roles/organizations, and some information around planned procurements (including a few we hope to be involved with).

So, whilst the public surface of SC16 might have been relatively sedate, the underworld of the HPC community was as seething with activity and information as ever.

Lies


Yes, lies. Or marketing for want of another name. Some of you may have observed me struggling to sustain my patience with the more outrageous marketing behavior that has bubbled through our HPC community recently. SC16 saw a renewed attack on our ability to withstand exaggerated and not-quite-misleading pitches of various HPC offerings.

In various fora over the last year or so I’ve said that sometimes I suspect GPU and cloud advocates are having a private competition to see who can get away with the most dodgy marketing claims. In the right cases, GPUs and cloud offer game-changing benefits to adopters. But neither of them are the solution to all the woes of computing that they each seem to claim.

GPUs don’t make overall workflows of a broad range of applications faster with straightforward effort. GPUs can be more expensive than CPUs when taking total costs into account. The full energy efficiency aspects of GPUs (application results per energy cost of the whole system) don’t always favor GPUs over CPUs.

Cloud is not cheaper or more flexible than a properly run in-house HPC facility. An in-house HPC facility can work with OpEx instead of CapEx just like cloud. It’s very rare (very) to find a HPC site that is running their facility at the under 40% utilization assumed by the most blatantly biased cloud marketing. (Actually, I do know of two sites with horrendously low utilization, but they are both special cases.)

These poorly judged marketing games annoy me not because I am anti-GPU or anti-cloud, but because I spend a fair proportion of my HPC consulting efforts helping my clients make sense of the real benefits of GPUs and cloud for them, and this isn’t helped by the degraded credibility created by the ambitious marketing claims.

Let’s not pretend GPU and cloud are the only culprits either. CPU companies, software vendors, HPC centers, and networking providers have all caused me to raise my eyebrows during the pre-SC16 build up and during SC16 itself.

And, of course, now that #BigData has been stapled to every event, product and HPC center, finally removing its differentiation for marketing, the community has latched onto a new must-be-used-for-all-marketing term: AI. AI is now the next big market, killer app, reason-our-product-is-better, solution to the world’s problems, must-tell story, etc.

Perhaps the only solution is for me to give in and join the extreme marketing game. Maybe I should just claim that our HPC consulting services will help you get 50% better capability for the same budget, or will migrate your code to new technologies and deliver 10x speedup within days. See – it makes you think it’s too good to be true, doesn’t it?

Trouble is, even in jest, I can’t make myself stray far from the truth. In fact, for specific projects, both of those claims have been true. But that doesn’t make them generally true – so what is fair game for our marketing?

My mild rant means I’ve waffled on too long now. The result for you the reader, if you’ve had the patience to get this far, is that part 2 of my SC16 wrap up (women and money) will follow in another blog post here: "Secrets, Lies, Women and Money: the definitive summary of SC16 - Part 2".

In fact, I've got enough material to write further posts, on specific issues such as HPE+SGI, Top500, how to make the most of SC16 (or SC17), and students. Need a new title for those parts though.

I hope I’ve not scared you off and look forward to you returning for more SC16 thoughts soon.


No comments: