On Thu, Sep 27, 2012 at 02:44:16PM -0700, John Buckman wrote: > The other area where "state of the art" thinking is occurring, is in > scaling web sites to many, many machines.
Reputedly, the best toolkit for building that sort of stuff is Erlang and its OTP libraries. (Asynchronous message passing, hot code reloading, fault tolerance, etc.) I've also long wondered if its Mnesia distributed RDBMS is any good. I was quite happy with AOLserver, Tcl, and a decent single-box RDBMS like Oracle or PostgreSQL. But if I intended to build stuff for massive scale out, ideally I'd want to first hack seriously with Erlang for a year or so first to really understand what it's good for, what it's not, clever approaches the Erlang community came up with, etc. If anybody here as done something along those lines and can report back, compare/contrast to the AOLserver / Tcl / OpenACS world, that could be really interesting! A maybe related approach from the "enterprise" software world (probably meaning giant investment banks), is Message-Oriented Middleware; this is also asynchronous message passing. This guy (Kirk Wylie, founder of OpenGamma) seems to know what he's talking about, and recommended the Hohpe book below: http://kirkwylie.blogspot.com/ http://www.amazon.com/Enterprise-Integration-Patterns-Designing-Deploying/dp/0321200683/ Enterprise Integration Patterns: Designing, Building, and Deploying Messaging Solutions by Gregor Hohpe, Bobby Woolf > Ousterhout recently wrote a paper about RAMCloud, which would be very helpful > on aolserver: > http://cacm.acm.org/magazines/2011/7/109885-the-case-for-ramcloud/fulltext Useful I guess, but that seems pretty low level. I'd rather look into shared-nothing parallel RDBMSs. These were mostly intended for analytic (date warehouse) loads, but there are OLTP-oriented designs available now too (e.g. VoltDB, which also happens to be RAM-only). I don't know how well they scale across modes. Years ago, I once heard that, Teradata was only intended to scale to no more than 100 or so fat nodes. But even if modern shared-nothing OLTP systems scale no better, that's still about 100 times better than you could do with a typical single server Oracle or PostgreSQL installation, which should give you a lot of helpful leeway before you HAVE to develop and adopt database sharding and/or more specialized tools, like the Digital City team did. The recent Calvin OLTP research is also interesting. Like "RamCloud", this is a lower-level tool, not a complete system. Unlike RamCloud though, it seems like an actual tested advance in the state-of-the art, not just a conceptual description of a cool piece of infrastructure someone might want to build: http://dbmsmusings.blogspot.com/2012/05/if-all-these-new-dbms-technologies-are.html -- Andrew Piskorski <a...@piskorski.com> ------------------------------------------------------------------------------ Got visibility? Most devs has no idea what their production app looks like. Find out how fast your code is with AppDynamics Lite. http://ad.doubleclick.net/clk;262219671;13503038;y? http://info.appdynamics.com/FreeJavaPerformanceDownload.html _______________________________________________ aolserver-talk mailing list aolserver-talk@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/aolserver-talk
