D. Mosberger, P. Druschel, and L. L. Peterson. Implementing Atomic Sequences on Uniprocessors Using Rollforward. In Software---Practice and Experience, 26(1), pp. 1-23, Jan. 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in the interrupt handler. It increases interrupt latency when the interrupt occurred inside the atomic procedure, because the procedure must first be finished. But the most severe downside of this technique is, that it inhibits inlining of memory accesses. Similar techniques are described in [9] [36], 35] 41] Figure 13a shows the change in performance when no revocation checks are performed. This configuration is slightly slower than a configuration that used the CLI method for revocation check. We can only explain this by code cache effects. Using spinlocks adds an additional overhead ....

D. Mosberger, P. Druschel, and L. L. Peterson. Implementing Atomic Sequences on Uniprocessors Using Rollforward. In Software---Practice and Experience, 26(1), pp. 1-23, Jan. 1996.

....access do not interfere with each other and interleave request arguments. Bypassing the operating 47 system when initiating I O requests avoids the cost of a system call and can lead to performance improvements especially for low latency I O operations such as network communication. Software [11][71] and hardware [47] 68] based synchronization mechanisms exist that implement user level atomic sequences, but these solutions always require that software voluntarily uses the synchronization mechanism to avoid corruption of shared data. Uncooperative applications can bypass the synchronization ....

D. Mosberger, P. Druschel, and L.L. Peterson, "Implementing Atomic Sequences on Uniprocessors Using Rollforward," Software---Practice and Experience, vol. 26, no. 1, Jan. 1996, pp. 1-24.

.... g1 clrb [ g1] Normally, the write succeeds, but when the poll page is write protected, the write results in a signal. The signal handler analyzes the signal to identify it as a poll, and then suspends the thread. Mosberger, Druschel, and Peterson refer to this technique as controlled faults [11]. The main advantage of this alternative poll sequence over the poll code shown in Section 2.3 is the reduction in the number of instructions from four executed (and six emitted) to two executed (and two emitted) Thus, in terms of time efficiency, the optimized poll code falls approximately ....

....the cost of garbage collection by approximately 5 4.7 = 23 . This is a significant speedup to get in a highly tuned memory system. 5 Related work Mosberger, Druschel, and Peterson discuss software implementations of atomic ( uninterruptible ) sequences for uniprocessor operating systems [11]. They advocate the use of roll forward: if a thread is about to be interrupted in the middle of an atomic sequence, it is rolled forward to the end of the sequence by the interrupt handler. In many regards, atomic sequences resemble the regions of code between GC points in our system. Mosberger, ....

David Mosberger, Peter Druschel, and Larry L. Peterson. Implementing Atomic Sequences on Uniprocessors Using Rollforward. Software---Practice and Experience, 26(1), p. 1-23, January 1996.

....the receiving thread to either suspend itself or throw an exception, as appropriate. The problem is that this may cause a thread to block (or even die) in the middle of a critical section, leaving the critical section locked. To eliminate this possibility Toba uses a limited form of roll forward [MDP96] to allow a thread interrupted by a signal to exit the critical section before handling the signal. Note that this problem also exists with critical sections in the Java code itself; the Java literature does not offer much of a solution other than recommending limited use of these asynchronous ....

David Mosberger, Peter Druschel, and Larry L. Peterson. Implementing atomic sequences on uniprocessors using rollforward. Software--- Practice and Experience, 26:1--23, 1996.

....the receiving thread to either suspend itself or throw an exception, as appropriate. The problem is that this may cause a thread to block (or even die) in the middle of a critical section, leaving the critical section locked. To eliminate this possibility Toba uses a limited form of roll forward [MDP96] to allow a thread interrupted by a signal to exit the critical section before handling the signal. Note that this problem also exists with critical sections in the Java code itself; the Java literature does not offer much of a solution other than recommending limited use of these asynchronous ....

David Mosberger, Peter Druschel, and Larry L. Peterson. Implementing atomic sequences on uniprocessors using rollforward. Software---Practice and Experience, 26:1--23, 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC