This 4D heatmap is showing node.js GCs (that is, V8 GCs) decomposed by latency and type. The program here is a simple one that generates garbage as quickly as possible. This shows several things: first, that node.js will GC frequently but for (relatively) short periods when garbage generation is high. (The high band is roughly 30 ms.) Second, it shows the relative performance of V8's GC algorithms: as expected, the scavenge GC (short-lived objects in the new generation) is quick, while the mark-sweep-compact GC (longer-lived objects in the older generation) is relatively slower. This data was gathered with DTrace via USDT probes in node.js, and visualized using a prototype of Joyent's forthcoming cloud analytics (on which more details can be found at
http://dtrace.org/blogs/brendan/2011/01/18/cloud-analytics-first-look/ ).