Distributed shared memory
In computer architecture, distributed shared memory (DSM) is a form of memory architecture where the (physically separate) memories can be addressed as one (logically shared) address space. Here, the term shared does not mean that there is a single centralized memory but shared essentially means that the address space is shared (same physical address on two processors refers to the same location in memory). Distributed Global Address Space (DGAS), is a similar term for a wide class of software and hardware implementations, in which each node of a cluster has access to shared memory in addition to each node's non-shared private memory.
A distributed-memory system (often called a multicomputer) consist of multiple independent processing nodes with local memory modules which is connected by a general interconnection network.Software DSM systems can be implemented in an operating system (OS), or as a programming library and can be thought of as extensions of the underlying virtual memory architecture. When implemented in the OS, such systems are transparent to the developer; which means that the underlying distributed memory is completely hidden from the users. In contrast, Software DSM systems implemented at the library or language level are not transparent and developers usually have to program differently. However, these systems offer a more portable approach to DSM system implementation.A Distributed shared memory system implements the shared-memory model on a physically distributed memory system.
Software DSM Implementation:
There are almost three ways of implementing a software distributed shared memory.
1) Page based which is Using the system’s virtual memory
2) Shared variable approach which is Using some routines to access shared variables
3) Object based : Ideally accessing to a shared data is through object oriented discipline.
Message passing vs DSM :
Software DSM systems also have the flexibility to organize the shared memory region in different ways. The page based approach organizes shared memory into pages of fixed size. In contrast, the object based approach organizes the shared memory region as an abstract space for storing shareable objects of variable sizes. Another commonly seen implementation uses a tuple space, in which the unit of sharing is a tuple.
Shared memory architecture may involve separating memory into shared parts distributed amongst nodes and main memory; or distributing all memory between nodes. A coherence protocol, chosen in accordance with a consistency model, maintains memory coherence.
Abstract view of DSM:
Advantages of DSM:
• System scalable
• Hides the message passing
• Can handle complex and large data bases without replication or sending the data to processes
• DSM is usually cheaper than using multiprocessor system
• No memory access bottleneck, as no single bus
• DSM provides large virtual memory space
• DSM programs portable as they use common DSM programming interface
• Shields programmer from sending or receive primitives
• DSM can (possibly) improve performance by speeding up data access.
Issues in Implementing DSM Software:
• Data is replicated or cached
• Reduce delays
• Semantics for concurrent access must be clearly specified
• DSM is controlled by memory management SW, OS, language run-time system
• Locating remote data
• Granularity: how much data is transferred in a single operation?
Disadvantages of DSM:
• Could cause a performance penalty
• Should provide for protection against simultaneous access to shared data such as lock.
• Performance of irregular problems could be difficult
Methods of Achieving DSM: There are usually two methods of achieving distributed shared memory 1)Hardware 2)Software
Hardware: such as cache coherence circuits and network interfaces
Software: We can use this method in different ways such as modifying the operating system kernel
Consistency Models: Memory system tries to behave based on certain rules in the system, which is called system’s “consistency model”
Memory coherence is one of the most important abilities into the system.It will make sure that a system executes memory operation correctly.Suppose we have n processes and Mi memory operation for each process Pi,and all the operations are executed sequentially.We can conclude that (M1 + M2 + . . .Mn)!/(M1!M2! . . .Mn!)are possible interleavings of the operations.The problem here is which of these interleavings are correct and which ones are wrong. The memory coherence defines which interleavings are permitted.Traditionally, Read operation returns the value written by the most recent Write which is ambiguous with replicas and concurrent accesses.
Strict Consistency: In strict consistency each operation is stamped with a global wall-clock time. Rules: – Rule 1: Each read gets the latest written value – Rule 2: All operations at one CPU are executed in order of their time stamps Multiprocessors maintain cache coherence by broadcasting writes to all processors, which can then either update or invalidate their local caches.
Since software DSM could not implement the atomic broadcasts needed to preserve strict consistency, other consistency models are needed.
Sequential Consistency(sc): Sequential consistency is specified as follows: If all operations of the processors were executed in sequential order,the result of any execution would be the same and each processor's operation appears in this sequence in the local program order.Interleaving of the operations which are coming from different processors is possible,but processors should meet the same interleaving. Sequential consistency is slightly weaker model than strict consistency.The most important difference between them is the sequential consistency does not assume real time. Rules: There exists a total ordering of operation such that : – Rule 1: Each machine’s own ops appear in order – Rule 2: All machines see results according to total order.For example: reads meet the most recent writes.In other words, the instructions of all processes are interleaved in some sequential order. For independent processes,sequential consistency presents no problem. For critical sections, there is a possibility of race conditions. Users who wish to enforce a certain order of execution could use synchronization mechanisms, exactly the same way in a shared memory processor.
Slow Memory consistency: Slow memory is one of the weakest consistency model which still are used for interprocess communication.In this consistency model all the processors must agree on the order of observed writes to each location by a single processor.Moreover, local writes should be visible instantly.
Replication in DSM:
Replication of shared data in general reduced network traffic,promotes increased parallelism, fewer page faults, and is more efficient than non-replicated implementations.
Main problem: preserving consistency when multiple copies are present.
Consistency in Structured DSM: Object-based or structured DSM may use more efficient consistency because it is easier to specify what is going to be shared.Users can identify points in the program where the data is consistent.They only share designated objects or variables.If shared data accesses happen only inside critical sections, while a process enters into a critical section, the DSM only needs to ensure that variables are consistent.
Two Consistency Models:
Release consistency: When a process exits a critical section, new values of the variables are propagated to all sites.
Entry consistency: When a process enters a critical section, it will automatically updates the values of the shared variables
Examples of such systems include:
- Patterson, David A. and John L. Hennessy (2007). Computer architecture : a quantitative approach, Fourth Edition, Morgan Kaufmann Publishers, p. 201. ISBN 0-12-370490-1.
- Distributed Shared Cache
- Memory coherence in shared virtual memory systems by Kai Li, Paul Hudak published in ACM Transactions on Computer Systems, Volume 7 Issue 4, Nov. 1989