A study of Java object demographics

Abstract

Researchers have long strived to exploit program behaviour in order to improve garbage collection efficiency. For example, by using a simple heuristic, generational GC manages short-lived objects well, although longer-lived objects will still be promoted to an older generation and may be processed repeatedly thereafter. In this paper, we provide a detailed study of Java object lifetimes which reveals a richer landscape than the generational view offers.Allocation site has been claimed to be a good predictor for object lifetime, but we show that object lifetime can be categorised more precisely than 'short-lived/long-lived/immortal'. We show that (i) sites allocate objects with lifetimes in only a small number of narrow ranges, and (ii) sites cluster strongly with respect to the lifetime distributions of the objects they allocate. Furthermore, (iii) these clusterings are robust against the size of the input given to the program and (iv) are likely to allocate objects that are live only in particular phases of the program's execution. Finally, we show that, in contrast to previous studies, (v) allocation site alone is not always sufficient as a predictor of object lifetime distribution but one further level of stack context suffices.

A black-box approach to understanding concurrency in DaCapo

“…Filtering out VM-specific activity is complicated by the VM allocating its objects in the same heap. We adopt the methodology of Jones & Ryder [16] and use Jikes RVM's baseline compiler. We count only operations by application threads, including the finaliser and reference handler threads; we exclude all other VM threads.…”

JVM-hosted languages

“…Dynamic studies of Java programs have generally focused on one particular aspect such as instruction mix [19] or object demographics [5,10]. Stephenson and Holst [28] present an analysis of multicodes, which are a generalization of bigrams [16] to variable length dynamic bytecode sequences.…”

Precise calling context encoding

“…Context sensitive optimizations [21,7] often specify how programs should behave in various contexts to achieve efficiency. For example, region-based memory management [7] tries to cluster memory allocations into large chunks, called regions, so that they can be explicitly managed; context sensitive region-based memory management specifies in which region an allocation should be performed under various contexts. Such analyses need to disambiguate the different contexts reaching a program point at runtime to decide if the current context is one of those specified.…”

“…From the encoding graph in Fig. 7 (b), we know that explicit IDs at I in range [4,7) represent contexts along the edge GI. We reverse both the stack offset and the explicit encoding along this edge and get of f set = 5 − 1 = 4 and id = 6 − 4 = 2.…”