System noise, OS clock ticks, and fine-grained parallel applications

Abstract

As parallel jobs get bigger in size and finer in granularity, "system noise" is increasingly becoming a problem. In fact, fine-grained jobs on clusters with thousands of SMP nodes run faster if a processor is intentionally left idle (per node), thus enabling a separation of "system noise" from the computation. Paying a cost in average processing speed at a node for the sake of eliminating occasional processes delays is (unfortunately) beneficial, as such delays are enormously magnified when one late process holds up thousands of peers with which it synchronizes.We provide a probabilistic argument showing that, under certain conditions, the effect of such noise is linearly proportional to the size of the cluster (as is often empirically observed). We then identify a major source of noise to be indirect overhead of periodic OS clock interrupts ("ticks"), that are used by all general-purpose OSs as a means of maintaining control. This is shown for various grain sizes, platforms, tick frequencies, and OSs. To eliminate such noise, we suggest replacing ticks with an alternative mechanism we call "smart timers". This turns out to also be in line with needs of desktop and mobile computing, increasing the chances of the suggested change to be accepted.

Understanding Timelines Within MPEG Standards

“…For example, Microsoft's Windows 7 OS offers a precision of 15.625ms [17], irrespective of the physical clock's resolution, and the current version of Minix Operative System (OS) offers a precision of 16ms [18]. Around 2004, although typical clock's resolution was around 10ms, the tendency was to improve systems clock's resolution in various OS, such as Linux, FreeBSD, DragonFlyBSD, up to 1ms [19]. Clock frequency is the rate at which a physical clock's oscillator fluctuates.…”

“…The values included in the 16-bit accessUnitDuration (AUduration) field and in the 16-bit compositionUnitDuration (CUduration) field are divided by the value of timescale to calculate the AU and CU time in seconds, as can be seen in eq. (19) and (20), respectively:…”

Balancing HPC applications through smart allocation of resources in MT processors

“…The OS noise has been recognized as one of the major source of extrinsic imbalancing [11], [25], [32]. A classical example is the interrupt annoyance problem present in Intel processors: all the interrupts coming from external devices are routed to CPU0, therefore, the OS noise caused by executing the interrupt handlers on CPU0 is higher than the noise on the other CPUs.…”

Active Flash: Out-of-core data analytics on flash storage

“…1 machine on Top500 in 1997 to 0.13 for Jaguar and 0.01 for the projected exaflop machine in 2018 [36], [47]. In addition, for a substantial class of HPC applications characterized by close, fine-grained synchronization between computation on different nodes, nondeterminacy resulting from competing CPU usage can result in severe performance impairment [48], as such jitter causes additional waiting for "stragglers" at each communication step.…”