Withdraw
Loading…
Palloc: parallel dynamic memory allocation
Simmons, Patrick A.
Loading…
Permalink
https://hdl.handle.net/2142/24135
Description
- Title
- Palloc: parallel dynamic memory allocation
- Author(s)
- Simmons, Patrick A.
- Issue Date
- 2011-05-25T15:07:09Z
- Director of Research (if dissertation) or Advisor (if thesis)
- Adve, Vikram S.
- Department of Study
- Computer Science
- Discipline
- Computer Science
- Degree Granting Institution
- University of Illinois at Urbana-Champaign
- Degree Name
- M.S.
- Degree Level
- Thesis
- Keyword(s)
- malloc
- memory allocation
- palloc
- palloc1
- palloc2
- dynamic memory allocation
- Abstract
- In this thesis, we describe two related memory allocators, each with novel properties. PALLOC1 contributes a unique strategy based on the traversal of a parallel tree data structure for allowing concurrent allocations and frees to proceed within a single thread's heap. PALLOC1 also provides a novel, provable guarantee limiting the allocator's requests for more memory from the operating system to only those situations where no contiguous block is available to satisfy the allocation request, and a pure bitmap allocation strategy speed-competitive even for sequential codes with the boundary-tag / binning strategy used in dlmalloc. We find that, for larger allocation patterns, our implementation exhibits competitive base performance relative to other parallel allocators, superior scaling, and better resistance in practice to fragmentation. PALLOC2 contributes a second unique strategy for memory allocation based on bitmap allocation into variable-sized superpages. Our system provides the runtime with the useful ability to, given an arbitrary heap address, find both the start of the heap allocation and the size of the object allocated. Thus, PALLOC2 provide the capabilities of baggy bounds checking with no performance impact. In fact, we find that, for both sequential and parallel programs, PALLOC2's performance is superior to PALLOC1 and to other state-of-the-art allocators including Hoard, DLMalloc, and Streamflow for allocations of all sizes.
- Graduation Semester
- 2011-05
- Permalink
- http://hdl.handle.net/2142/24135
- Copyright and License Information
- Copyright 2011 Patrick A. Simmons
Owning Collections
Dissertations and Theses - Computer Science
Dissertations and Theses from the Dept. of Computer ScienceGraduate Dissertations and Theses at Illinois PRIMARY
Graduate Theses and Dissertations at IllinoisManage Files
Loading…
Edit Collection Membership
Loading…
Edit Metadata
Loading…
Edit Properties
Loading…
Embargoes
Loading…