C4

Abstract

C4, the Continuously Concurrent Compacting Collector, an updated generational form of the Pauseless GC Algorithm [7], is introduced and described, along with details of its implementation on modern X86 hardware. It uses a read barrier to support concurrent compaction, concurrent remapping, and concurrent incremental update tracing. C4 differentiates itself from other generational garbage collectors by supporting simultaneous-generational concurrency: the different generations are collected using concurrent (non stop-the-world) mechanisms that can be simultaneously and independently active. C4 is able to continuously perform concurrent young generation collections, even during long periods of concurrent full heap collection, allowing C4 to sustain high allocation rates and maintain the efficiency typical to generational collectors, without sacrificing response times or reverting to stop-the-world operation. Azul systems has been shipping a commercial implementation of the Pauseless GC mechanism, since 2005. Three successive generations of Azul's Vega series systems relied on custom multi-core processors and a custom OS kernel to deliver both the scale and features needed to support Pauseless GC. In 2010, Azul released its first software-only commercial implementation of C4 for modern commodity X86 hardware, using Linux kernel enhancements to support the required feature set. We discuss implementation details of C4 on X86, including the Linux virtual and physical memory management enhancements that were used to support the high rate of virtual memory operations required for sustained pauseless operation. We discuss updates to the collector's management of the heap for efficient generational collection and provide throughput and pause time data while running sustained workloads.

Scalable concurrent and parallel mark

“…Also, since objects live and die in clusters [22], mark-bits can be tested and cleared in groups. For these reasons, mark-bitmaps are prevalent in implementations of most commercial garbage collection algorithms [12,13,28].…”

“…This is primarily attributable to the poor temporal and spatial locality properties of a pointer-chasing algorithm. Table 1, shows the variation in the percentage of GCcycle time spent on marking with increasing number of task-threads for the C4 [28] algorithm. The numbers reflect average markphase duty-cycles for the Dacapo benchmark suite [8] for the new generation with the old generation displaying a similar pattern.…”

“…The proposed scheme is a first of its kind "hybrid" work-sharing algorithm that tries to reduce cache traffic while keeping all involved processors busy. Mutator threads are the other source of tracing work in concurrent [11,28] as well as on-the-fly marking schemes [6,15]. In these schemes, mutator threads are provided with thread-local marking stacks where they collect references to potentially "gray" [14] objects.…”

“…Azatchi [6] uses soft handshakes to drain this work. The C4 collector [28] uses soft handshakes, referred to as checkpoints, to transfer this work to a global pool of marking stacks; the collector's taskthreads then acquire these stacks from the global pool in a threadsafe manner and mark through the references in these stacks.…”

“…Our proposed algorithms have been prototyped within the C4 collector [28], that runs on a Linux based X86 machine. We would like to point out that the proposed algorithms are independent of the C4 collector algorithm and can be applied to other mark-sweep or mark-compact type GC algorithms.…”