PMAKEParallel Make (Parallel Computation)
Copyright 1988-2018, All rights reserved.
References in periodicals archive ?
We injected faults while the system was executing the pmake workload, waited for the system to detect and recover from the fault, then checked all output files against reference copies to ensure that any wild writes due to the fault had not corrupted application output.
The reduction in Pmake is primarily due to the monitor initializing pages on behalf of the kernel and hence suffering the memory stall and instruction execution overhead of this operation.
Table II shows the overhead of the virtualization separately for the top OS services for the Pmake workload.
We therefore ran the Pmake workload in this new configuration in SimOS both to predict the performance of the "real" hardware and to compare it with the IRIX 5.3 simulation results.
For the same Pmake workload, the system time "only" increases by a factor of 1.62, whereas it increased by 2.15 in the 32-bit kernel.
The workload consists of eight different copies of the basic Pmake workload.
The first workload is the Pmake workload and uses the six configurations of Section 5.3.
All configurations use the same physical resources - eight processors and 256MB of memory - but the Pmake workload uses different virtual configurations.
We see in Figure 8 that IRIX suffers from high synchronization and memory system overheads for the Pmake workload, even at just eight processors.
Using a single virtual machine leads to higher overheads than in the comparable uniprocessor Pmake workload.
Origin200 Execution Time Pmake Engineering Breakdown IRIX Disco Ratio IRIX Disco Ratio (sec.) (sec.) (sec.) (sec.) User 11.3 11.7 1.03 65.2 69.7 1.07 Kernel 5.9 9.6 1.62 0.2 0.2 1.00 Idle 13.1 11.4 0.87 0 0 Total 30.3 32.7 1.08 65.4 69.9 1.07 6.2 Overheads of Virtualization
(1) Pmake: This workload compiles Disco itself using the SGI development tools, two files at a time.