Check out the new USENIX Web site. next up previous
Next: Memory Fault in Data Up: Experimental Results Previous: Self-Healing

Memory Fault in Stack Pointer

SSM heals itself in the presence of a memory fault; performance and throughput is unaffected, and SSM recovers from the fault.

Using ptrace(), we monitor a child process and change its memory contents. In this benchmark, $W=3, WQ=2, R=2$, data size is 8KB, and we increase t to 100ms to account for the slowdown of bricks using ptrace. Figure 10 shows the results of injecting a bitflip in the area of physical memory where the stack pointer is held. The fault is injected at time 14; the brick crashes immediately. The lightened section of figure 10 (time 14-23) is the time during which only five bricks are running. At time 23, Pinpoint detects that the brick has stopped sending heartbeats and should be restarted, and restarts the brick; the system tolerates the fault and successfully recovers from it.

Figure: Fault Injection: Memory Bitflip in Stack Pointer.



Benjamin Chan-Bin Ling 2004-03-04