As the latency of the network approaches that of memory, it becomes increasingly attractive for applications to use remote memory---random-access memory at another computer that is accessed using the virtual memory subsystem. This is an old idea whose time has come, in the age of fast networks. To work effectively, remote memory must address many technical challenges. In this paper, we enumerate these challenges, discuss their feasibility, explain how some of them are addressed by recent work, and indicate other promising ways to tackle them. Some challenges remain as open problems, while others deserve more study. In this paper, we hope to provide a broad research agenda around this topic, by proposing more problems than solutions.

Transmitting messages in mobile wireless networks typically involves on-demand route discovery implemented via network-wide broadcast. Due to the dynamic nature of the network topology the life-time of a route is very short, so a source frequently requires a new route to an old destination. Simultaneous discovery of multiple routes can reduce the overhead due to repeated route discovery broadcasts. Previously proposed multipath protocols do not guarantee discovery of alternative paths if they exist. We propose a multiple route discovery algorithm (ALTDSR) that finds a (multihop) primary path between a source and a destination, and a set of alternative paths. We introduce dominator relationships between primary and non-primary path nodes. Using dominators, we characterize alternative paths that bypass an intermediate node on the primary path. We develop algorithms that guarantee finding a set of alternative paths to tolerate any single node fault on the primary path, if such a set of alternative paths exists. We present simulation results which show that under high mobility conditions, ALTDSR delivers substantially more packets (around 75% more) than dynamic source routing (DSR) with moderate increase in routing overhead.

A control flow fault occurs when a processor fetches and executes an incorrect next instruction. Executable assertions, i.e., special instructions that check some invariant properties of a program, provide a powerful and low-cost method for on-line detection of hardware-induced control flow faults. We propose a technique called ACFC (Assertions for Control Flow Checking) that assigns an execution parity to a basic block, and uses the parity bit to detect faults. Using a graph model of a program, we classify control flow faults into skip, re-execute and multi-path faults. We derive some necessary conditions for these faults to manifest themselves as execution parity errors. To force a control flow fault to excite a parity error, the target program is instrumented with additional instructions. Special assertions are inserted to detect such parity errors. We have a developed a preprocessor that takes a C program as input and inserts ACFC assertions automatically. We have implemented a software-based fault injection tool SFIG which takes advantage of the GNU debugger. Fault injection experiments show that ACFC incurs less performance overhead (around 47%) and memory overhead (around 30%) than previous techniques, with no significant loss in fault coverage.