Diehard allocator - custom allocators, Operating System

Allocators can also be used to avoid problems with unsafe languages. C and C++ are pervasive, with huge amounts of existing code. They are also memory-unsafe languages, in that they allow many errors and security vulnerabilities. Some examples include double free(), invalid free(), uninitialized reads, dangling pointers, and buffer over?ows in both stack and heap buffers.

DieHard is an allocator developed at UMass which provides (or at least improves) soundness for erroneous programs. There are several hardware trends which are occurring: multicore processors are becoming the norm, physical memory is relatively inexpensive, and 64-bit architectures are increasingly common, with huge virtual address spaces. Meanwhile, most programs have trouble making full use of multiple processors. The net result is that there may soon be unused processing power and enormous virtual address spaces.

If you had an in?nite address space, you wouldn't have to worry about freeing objects. That would mostly eliminate the double free(), invalid free(), and dangling pointer bugs. And if your heap objects were in?nitely far apart in memory, you wouldn't need to worry about buffer over?ows in heap objects.

DieHard tries to provide something along these lines, within the constraints of ?nite physical memory. It uses randomized heap allocation, so objects are not necessarily contiguous in virtual memory. Since the address space is actually ?nite, objects won't actually be in?nitely far apart, and buffer overruns might actually cause collisions between heap objects. But this is where the multicore processors come in: With the unused processor cores, run multiple copies of the application, say three copies, each allocating into their own randomized heap. So the heap errors are independent among the three copies of the application. All copies get the same input, and the output is the result of voting among the three copies of the program. If one instance of the application disagrees with the other two, it is killed, since there was likely a collision between heap objects in that one. Similarly, if one instance dies with a segfault or other error, the others remain running.

Surviving copies can be forked to replace copies which were killed off, though this reduces the independence among copies.

Posted Date: 3/13/2013 2:49:51 AM | Location : United States







Related Discussions:- Diehard allocator - custom allocators, Assignment Help, Ask Question on Diehard allocator - custom allocators, Get Answer, Expert's Help, Diehard allocator - custom allocators Discussions

Write discussion on Diehard allocator - custom allocators
Your posts are moderated
Related Questions
Q. What is the occupation of the object manager? Answer: Substance present a generic set of kernel mode interfaces to user mode programs. Objects are influenced by the execut

Question : a) Describe the difference between the authentication of a user account on a peer to peer network and a client server network. b) Compare an end user account with

Briefly discuss on I/O hardware? A controller is a collection of electronics that can operate a port, a bus, or a device. A serial-port controller is an easy controller. It is

Acyclic-graph directories An acyclic graph permits directories to have shared subdirectories and files. The similar file or subdirectory may be in two dissimilar directories. A

Define a loader that is executed when a system is first turned on or restarted? Bootstrap loader is executed when a system is first turned on or restarted

What are batch systems? Batch systems are quite appropriate for implementing large jobs that need little interaction. The user can submit jobs and return later for the results.

what are the main advantages of multiprogramming

what do obelia eat and how do they feed

Contiguous allocation The contiguous-allocation method needs each file to occupy a set of contiguous blocks of the disk. Disk addresses explain a linear ordering on the disk. T

Developing Code with Threads Writing code to support a thread-based implementation is very straightforward. Simple APIs are presented to the developer for each threads package.