Paper author here, thanks for the interest :). I just wanted to elaborate on the overhead numbers you mentioned.
I am not specifically familiar with the 35%+ numbers, but you should take into account the cost of any defense on top of the cost of putting tags in pointers, because that makes all the difference.
We found in our experiments on SPEC CPU2006 that, without static analysis for optimizations, the cost of pointer tagging itself (tagging + masking) is 22% geomean, which is almost entirely due to masking on loads/stores. This is with a 64-bit masking constant (0x80000000ffffffff) which is unique to our approach and is relatively inefficient on x86. With 32-bit masking (0xffffffff constant), which is more common, the number is about 14%.
This only covers metadata management and is not useful on its own, so you need to implement some kind of defense on top of this which is usually costly. The hardware support used by HWAsan only removes masking overhead by having the (ARM) hardware ignore the upper byte of the pointer, which is a nice way to not having to pay for masking.
In "Parameter Selection" you say "We simplify this by having service owners define the current number of outliers, if there are any, at configuration time."
How do service owners get this data? by manual inspection of graphs like the ones shown earlier?
I have the same issue on Chrome 40.0.2214.111 on Linux. Other than that I agree, it's nice to have multiple alternatives for different purposes. For example, there is also Skeleton (http://getskeleton.com/) which has a very minimalistic design as well.
The examples are very relatable, at least for me personally.
Judging from this fragment, I would consider myself a stack person:
"For tasks that represent pain that must be immediately alleviated, it’s not that the stack feels I must fix it; rather, this must be fixed. The task is painful not because the stack must fix it, but because it exists, period."
The frustrating part about this is that it is very hard to explain to coworkers why you feel this, because you intuitively expect them to feel the same way. I can recall some cases in which this has caused some friction when working on group projects.
By the way, I do think pattern matching could have been discussed more extensively. Pattern matching is one of the reasons I love functional programming so much, since combined with algebraic types they can be much more expressive than imperative control-flow statements (e.g. if-else). For programmers that are used to imperative programming, discovering the power of pattern matching can be quite a hurdle.
A good read, I especially like the parts about error handling and maps/applicatives. They are well-illustrated and tackle some difficult common problems I had shortly after I started using functional languages.
The part about functors/monads/monoids is also nice, although I feel like it would be better with the accompanying talk to bind it together a bit more.
Nice explanation, the overall layout and text-to-code ratio make it easy/fun to follow. I would like to see the finished post when all the TODO's are filled in. I would suggest to add "AMD64" explicitly to the page title for clarity.
So, I just completed the "curiosity killed the mouse" level, took 3 people a while to realise what was going on, but after a few minutes we got to the non-obvious solution, and I got sent back to the first level. I feel a bit screwed over now... =|
A difference is that a GPU uses a lot of power and takes up a lot of space. I can imagine an optimized, energy-efficient chip would be useful in embedded systems. Something like a Raspberry Pi for image processing maybe?