Remote memory

Cui, Z. Z., Wang, H., Tan, Y. et al. Inducible and reversible NR1 knockout reveals crucial role of the NMDA receptor in preserving remote memories in the brain. Neuron 41 781-793, 2004. 

Memory Complex mental function having four distinct phases (1) memorizing or learning, (2) retention, (3) recall, and (4) recognition. Clinically, it is usually subdivided into immediate, recent, and remote memory. [NIH]... 

Memory deficits are typically the first to appear in Alzheimer s disease. There is a pronounced impairment of explicit long-term memory. This involves subjective memory for events (episodic) and for factual information (semantic). Remote memories are impaired, but less so than recent ones. 

Amygdala paralimbic cortex + Emotion I Remote memory... 

Leplow B, Dierks C, Herrmann P, Pieper N, Annecke R, Ulm G (1997) Remote memory in Parkinson s disease and senile dementia. Neuropsychologia 35 547-555 Lesne S, Koh MT, Kotilinek L, Kayed R, Glabe CG, Yang A, Gallagher M, Ashe KH (2006) A specific amyloid-beta protein assembly in the brain impairs memory. Nature 440(7082) 352-357... 

NUMA is one of the fundamental concepts needed to understand the design of a parallel software application. Every modern computer has several levels of memory, and parallel computers tend to have more levels than uniprocessors. Typical memory levels in a parallel computer include the processor registers, local cache memory, local main memory, and remote memory. If the parallel computer supports virtual memory, local and remote disk are added to this hierarchy. These levels vary in size, speed, and method of access. In this chapter, we will lump all these differences under the general term nonuniform memory access (NUMA). Note that this is a broader use of the term than is often found in computer science literature, where NUMA often refers only to differences in the speed with which given memory items can be accessed using the same method. In our use, memory access is often synonymous with data transfer. ... 

Memory access costs are determined by the interaction between program structure and the performance characteristics of the computer system. Understanding these interactions at a high level of detail is typically complicated. However, a useful rough approximation of the memory access cost for a program can be obtained by modeling each transfer between memory levels as a fixed start-up cost (S) plus an incremental transfer cost per data unit (X). For example, the cost for a processor to fetch L data units at once from a remote memory into its local memory, cache, and registers can be modeled as follows ... 

Tolerance of latency and low bandwidth for references to remote memory locations... 

Here the execution time (T) of the program is described as a function of the number of processors ( ), the problem size (N), the memory size (M), the latency and transmission time of remote memory references (fO,tl), and other relevant parameters. We understand already that T can be a nonlinear function of the problem size and sensitive to the algorithm [e.g., conventional self-consistent field (SCF),29 T(N) = 0 N ) + 0 N ) vs. pseudospectral SCF," T(N) = 0 N ) + 0(N3)], and similarly the dependence on P can be complex. As discussed earlier, one measure of the efficiency of parallel execution may then be derived as follows ... 

Data Reuse The use of data fetched from a remote location multiple times. The practice offsets the initial cost of a remote memory fetch. 

Figure 3 The multicomputer. The nodes in the cluster are Eckert-von Neumann processors. Nodes may be of varying processing power. Nodes communicate via a network of some kind. The cost of sending a message between two processors is only a function of the size of the message and does not depend on the relative node locations and other network traffic. The memories of the nodes are private. The node local memory is faster to access than remote memory.

The memory hierarchy in a hypothetical MIMD machine constructed from the nodes in Figure 2.15 and the network in Figure 2.12. Logically equivalent data location types are subdivided by how many network hops are needed to reach them, Whop (intra-node hops for local memory, inter-node hops for remote memory). The estimate of the time required to access memory at the given location is access (this time is hypothetical and not based on any particular hardware). The level of treatment typically used for each memory level is given for the several layers of software... 

Various libraries providing differing levels of support for one-sided communication operations are currently available for parallel application development. Notably, the Aggregate Remote Memory Copy Interface, ARMCI,... 

The Aggregate Remote Memory Copy Interface, ARMCI, is available on the World Wide Web at http //www.emsl.pnl.gov/docs/parsoft/armci/index. html. 

---