Report of ISMM'02, Berlin

• Invited talk: The Business Importance of Java Garbage Collection, Robert F. Berry
  Robert explained how efficient memory mangement was becoming more and more important to companies like IBM. He said the IBM were still unconvinced of that generational GC was an effective solution. His customers demanded low worst-case pause times so the IBM production collector avoided copying. He raised a number of challenges for the community. How to manage the next generation of multi-GB heaps?
  

   

• Applying Priorities to Memory Allocation, Sven Robertz
  Robertz showed how to apply priorities to memory allocation in order to enhance the robustness of real-time applications. The proposed mechanisms can also be used to increase performance of systems with automatic memory management by limiting the amount of garbage collection work.A way of introducing priorities for memory allocation in a Java system without making any changes to the syntax of the language is proposed and this has been implemented in an experimental Java virtual machine and verified in an automatic control application.
  

   

• Reducing the Pause Time of Conservative Collectors, Toshio Endo, Kenjiro Taura and Akinori Yonezawa
  This paper described an incremental conservative garbage collectorc. Similar to Boehm et al's collector, it uses the OS's page protection mechanism for write barriers. They bound the number of dirty pages during concurrent marking in order to get maximum pause times of less than 5ms on 400MHz UltraSPARC processors although throughput is 1.2--53% higher than a stop-the-world collector.
  

   

• Automated Discovery of Scoped Memory Regions for Real-Time Java, Morgan Deters and Ron Cytron

   

• Understanding the Connectivity of Heap Objects, Martin Hirzel, Johannes Henkel, Amer Diwan and Michael Hind
  The idea is to explore the connectivity of objects in order to partition the graph. They claim that connectivity correlates strongly with object lifetimes and deathtimes and is therefore likely to be useful for partitioning objects.
  

   

• Visualising the Train Garbage Collector, Tony Printezis and Alex Garthwaite
  Tony showed how GCspy could visualise the Train collector. Visualisation revealed previously unexpected patterns of behaviour (very long trains, very popular objects etc)
  

   

• Estimating the Impact of Heap Liveness Information on Space Consumption in Java, Ran Shaham, Elliot Kolodner and Mooly Sagiv
  Shaham compared the difference in time between an object is collected by a GC and the earliest time it could be collected, given perfect knowledge. They obtained an average potential savings of 39% using an interface with complete liveness information, or 15% using a more practical approach.
  

   

• Thread-Local Heaps for Java, Tamar Domani, Gal Goldstein, ElliotKolodner, Ethan Lewis, Erez Petrank and Dafna Sheinwald
  The idea is, like Andy King's, to be able to collect objects belonging to separate threads independently, without stopping the whole world. Unlike Andy, Domani et al use a purely runtime approach (no static analysis). Whenever a thread is about to copy a reference to a local object A into a field of a shared object B, they first mark the transitive closure of A as shared.
  

   

• Heap Architectures for Concurrent Languages using Message Passing, Erik Johansson, Konstantinos Sagonas and Jesper Wilhelmsson
  This paper discussed how interprocess communication and garbage collection for the Erlang/OTP system.
  

   

• An Algorithm for Parallel Incremental Compaction, Ori Ben-Yitzhak and Victor Leikehman
  The authors avoid long pause-times for compacting the heap by choosing a small region of the heap and compacting only that.
  

   

• Using Passive Object Garbage Collection Algorithms, Abhay Vardhan and Gul Agha
  An elegant talk showing how 'normal' GC techniques can be applied to actor systems. A difficulty with actors is that an actor is alive if either another active has its address or the actor has the address of a live actor (because it could then send this actor its own address). Gulp!
  

   

• Adaptive Caching for Demand Prepaging, Scott Kaplan, Lyle McGeoch and Megan Cole
  The authors examined prepaged allocation and its interaction with two other important demand prepaging parameters: the degree, which is the number of extra pages that may be fetched at each page fault, and the predictor that selects which pages to prepage.
  

   

• An Adaptive Region-based Allocator for Java, Feng Qian and Laurie Hendren
  The authors link a region with each stack frame. Initially they assume no objects escape and so can be placed in their frame's region. Objects for which references escape their frame's scope, are trapped by a write barrier. Further allocations at a site which allocated an escaped object are allocated in a global region. I wasn't totally convinced by this approach.
  

   

• Dynamic Memory Management for Programmable Devices, Sanjeeve Kumar and Kai Li
  

   

• Accurate Garbage Collection in an Uncooperative Envirnoment, Fergus Henderson
  The paper described how to do fully type-accurate garbage collection in an uncooperative environment, using a "shadow stack" to link structs of pointer-containing variables, together with the data or code needed to trace them. It was implemented for the Mercury Prolog system.
  

   

• Sofware Caching vs. Prefetching, Aneesh Aggarwal
  

   

• Mostly Lock-Free Malloc, Dave Dice and Alex Garthwaite
  A well-delivered talk on Multi-Processor Restartable Critical Sections, or MP-RCS. MP-RCS permits user-level threads to know precisely which processor they are executing on and then to safely manipulate CPU-specific data, such as malloc metadata, without locks or atomic instructions. MP-RCS avoids interference by using upcalls to notify user-level threads when preemption or migration has occurred.

   

• In or Out? Putting Write Barriers in their Place, Steve Blackburn and Kathryn McKinley
  Blackburn gets performance benefits in the JikesRVM system by inlining the fast-path of write-barriers and using calls for the slow path.

Problems with this page?
Contact the mm-net webmaster
Last modified Fri Jan 28 18:18:01 GMT 2005