The subject of scalability becomes the topic considered by Stefan Hinker in his blog entitled Some thoughts about scalability in which he reviews the current thinking on the issue. The blog described the "hockey stick" curve that plots degradation of response time in terms of load. As Hinker notes, "a new product usually has a few large bottlenecks, and because they're large, the response time curve take off for infinity early, and goes almost straight up. Overloading the system even a little bit causes it to "hit the wall" and seem to hang. Response time is fairly flat until an inflection point, then heads up like a homesick angel."
While PCs have a tendency to reach the inflection point relatively rapidly when load exceeds 70 to 80%, the fine-grained spin-locking in Solaris typically delivers superior scaling, Hinker asserts. Other attributes of Solaris, such as the large set of evolved architecturally-aware features that make the Solaris scheduler itself a huge factor in the superior scaling of Solaris, along with such other features as evolved AIO options and preemption control, which have been well-integrated.
Hinker suggests that his readers use the Java2D demo found in the JDK and then open the dummy CPU loads and push the %CPU higher and higher until it finally starts to feel slow which, amazingly, he writes, often happens quite close to 100% before any actual, user-visible degradation shows up.
He suggests an even more graphic demonstration that uses SRM to grant an app 80% of the cpu and then start dozens of dummy CPU loads in another zone to force the CPU to pin at 100%, while performance stays fine. It may be difficult to convince an audience that the results are not being faked, he contends.
How Hybrid Storage Pools Provide Greater Efficiency for Less
Read More ...