Disclosure: Kayvon and I were office mates, and I've supported his research while at Google (and the wchrichton's research as well).
Amusingly, most of the papers highlighted come from industry experience / application. Part of this is self selection (Kayvon knows these people and these papers). But I believe an unstated component is one of the challenge of academic systems research: access to the seriously interesting challenges. Thus isn't specific to graphics, as many of my favorite papers from ISCA, NSDI, etc. are also industry-led backwards looking systems papers (though many, if not most, of the authors previously went to grad school).
Industry/Academia partnerships often produce the best systems work, precisely because industry can offer the challenge, while academics can step back and see how they might solve the problem completely differently. The natural tendency in most companies is to keep stringing along some small improvements that last you another six months. While it's true that every once in a while, someone does a big rewrite, they are often rare and for good reason. My favorite example came from a talk Jeff Dean gave at Stanford, where he explained you should design a system to scale 10x but not 100x, because by then it should be redone.
I was wondering how SIGGRAPH's "sexiness" paper criterion played out with all consequences, but never really sat down and looked into it, being satisfied with all the highlights and cool results.
Rejecting "systems papers" despite being valuable and novel definitely seems like an oversight this could produce. As "flashiness" is a key part of SIGGRAPH, I'm wondering whether there could be an alternative conference, or maybe a subconference that's as competitive/prestigious but is a bit more friendly to stuff like graphics systems?
Actually, that was one of the goals of High Performance Graphics. We explicitly merged the low brow Ray Tracing Symposium and Graphics Hardware, precisely so we could publish and discuss interesting systems-ey things that move the world forward but often aren't really broad or sexy enough for a general SIGGRAPH audience. I don't think there's a great way for it to be considered equally prestigious, but as an author you know (for better or for worse) that you'll have more sympathetic readership.
Amusingly, most of the papers highlighted come from industry experience / application. Part of this is self selection (Kayvon knows these people and these papers). But I believe an unstated component is one of the challenge of academic systems research: access to the seriously interesting challenges. Thus isn't specific to graphics, as many of my favorite papers from ISCA, NSDI, etc. are also industry-led backwards looking systems papers (though many, if not most, of the authors previously went to grad school).
Industry/Academia partnerships often produce the best systems work, precisely because industry can offer the challenge, while academics can step back and see how they might solve the problem completely differently. The natural tendency in most companies is to keep stringing along some small improvements that last you another six months. While it's true that every once in a while, someone does a big rewrite, they are often rare and for good reason. My favorite example came from a talk Jeff Dean gave at Stanford, where he explained you should design a system to scale 10x but not 100x, because by then it should be redone.