4 ms·

Optimizing Indirect Memory References with milk

more_original 10y ago
These press releases are terrible. It took me quite a while to find the original paper: Optimizing Indirect Memory References with milk http://dl.acm.org/citation.cfm?id=2967948 http://dl.acm.org/citation.cfm?id=2967948 (seems to be available without paywall)
- forgotpwtomain 10y ago
  > These press releases are terrible. It took me quite a while to find the original paper: Was about to post the same-thing, how do these articles get past the editor? At any-rate it may be worth switching to the ACM publication link, the mit-news page is totally uninformative. [Optimizing Cache Performance for Graph Analytics](https://arxiv.org/abs/1608.01362 https://arxiv.org/abs/1608.01362) also looks to be relevant (same ,authors Aug 2016).
- sctb 10y ago
  Thanks, we updated the link from http://news.mit.edu/2016/faster-parallel-computing-big-data-0913 http://news.mit.edu/2016/faster-parallel-computing-big-data-....
tempodox 10y ago
An error occurred while processing your request. Reference #50.bfc0c16d.1473861325.1484741f I guess the milk went stale.
- _delirium 10y ago
  It looks like the link here is trying to deep-link the PDF, which will work if you've already visited it "properly", but not if you haven't (haven't dug into what precisely it's checking, maybe a cookie, or your IP). You have to click through from here: http://dl.acm.org/citation.cfm?id=2967948 http://dl.acm.org/citation.cfm?id=2967948
- unsignedqword 10y ago
  Interesting, most people would say "the milk went sour" or "the milk soured," and not that it went "stale." "Stale" as a word just sounds weird in the context of a drink.
convolvatron 10y ago
the press release is terrible. the paper is fine. this is really interesting work. of course the paper gives a much more nuanced view of the performance benefits. this is effectively scheduling to hide memory latency (ala SMT) except done entirely dynamically, with batching to support locality. the most impressive part of the paper isn't just the speedup, but the speedup in the presence of the substantial overhead involved in capturing, scheduling, and restoring the closures. it would be nice if the super-fine-grained threading model made more than the occasional appearance. it opens up a lot of potential for hiding all kinds of latencies. if a runtime can reap substantial benefits, imagine what an architecture could do (i.e. SMT)

These press releases are terrible. It took me quite a while to find the original paper: Optimizing Indirect Memory References with milk http://dl.acm.org/citation.cfm?id=2967948 http://dl.acm.org/citation.cfm?id=2967948 (seems to be available without paywall)

forgotpwtomain 10y ago
> These press releases are terrible. It took me quite a while to find the original paper: Was about to post the same-thing, how do these articles get past the editor? At any-rate it may be worth switching to the ACM publication link, the mit-news page is totally uninformative. [Optimizing Cache Performance for Graph Analytics](https://arxiv.org/abs/1608.01362 https://arxiv.org/abs/1608.01362) also looks to be relevant (same ,authors Aug 2016).
sctb 10y ago
Thanks, we updated the link from http://news.mit.edu/2016/faster-parallel-computing-big-data-0913 http://news.mit.edu/2016/faster-parallel-computing-big-data-....

tempodox 10y ago

An error occurred while processing your request. Reference #50.bfc0c16d.1473861325.1484741f I guess the milk went stale.

_delirium 10y ago
It looks like the link here is trying to deep-link the PDF, which will work if you've already visited it "properly", but not if you haven't (haven't dug into what precisely it's checking, maybe a cookie, or your IP). You have to click through from here: http://dl.acm.org/citation.cfm?id=2967948 http://dl.acm.org/citation.cfm?id=2967948
unsignedqword 10y ago
Interesting, most people would say "the milk went sour" or "the milk soured," and not that it went "stale." "Stale" as a word just sounds weird in the context of a drink.

convolvatron 10y ago

the press release is terrible. the paper is fine. this is really interesting work. of course the paper gives a much more nuanced view of the performance benefits. this is effectively scheduling to hide memory latency (ala SMT) except done entirely dynamically, with batching to support locality. the most impressive part of the paper isn't just the speedup, but the speedup in the presence of the substantial overhead involved in capturing, scheduling, and restoring the closures. it would be nice if the super-fine-grained threading model made more than the occasional appearance. it opens up a lot of potential for hiding all kinds of latencies. if a runtime can reap substantial benefits, imagine what an architecture could do (i.e. SMT)