# HG changeset patch # User Noah Evans # Date 1307854809 0 # Node ID 76d5b329699ee8f4b2488bc5f84e71f273584a4d # Parent 66071a74d51ae482f43b065eceb5a43a27cbc0fc daily push: Sat Jun 11 07:00:04 CET 2011 diff -r 66071a74d51a -r 76d5b329699e doc/papers/nix/bltj.ms --- a/doc/papers/nix/bltj.ms Sat Jun 11 15:57:44 2011 +0000 +++ b/doc/papers/nix/bltj.ms Sun Jun 12 05:00:09 2011 +0000 @@ -18,8 +18,8 @@ running on dedicated hardware. Virtualisation enables the dynamic provisioning of resources with demand, and the possibilities of entire system backup, migration, and persistence; for many tasks, the -flexibility of this approach, coupled with its cost, management, and -energy reductions, are compelling. +flexibility of this approach, coupled with savings in cost, management, +and energy, are compelling. .PP However, certain types of tasks — such as compute-intensive parallel computations — are not trivial to implement as cloud applications. @@ -27,7 +27,7 @@ where synchronization and work allocation happen at fixed intervals. Interrupts from the hypervisor and the operating system add “noise” and, thereby, random latency to tasks, slowing down all the other -tasks by making them wait such slowed-down tasks finish. The effect +tasks by making them wait for such slowed-down tasks to finish. The effect may cascade and seriously disrupt the regular flow of a computation. .PP To avoid latency issues, HPC tasks are performed on heavily customized