[SSI-devel] changes to "top" on OpenSSI
Brought to you by:
brucewalker,
rogertsang
From: Roopa P. <pra...@in...> - 2004-10-14 10:36:17
|
Hi all, Sometime back we had a discussion on this list for changes that are required for "top" on OpenSSI. I am planning to implement a few of them. Below is the understanding of what changes to "top" are required. "top" on ssi currently displays a mixture of clusterwide and local data. What is local info and what is not in existing "top" output: ------------------------------------------------------------ 1) Header section : uptime : local load average : local processes (sleeping, running, zombie and stopped) : clusterwide CPU states: local Mem : local Swap : local 2) Processes : Shows clusterwide processes Clusterwide "top" for SSI: -------------------------- 1) Default usage : This would display clusterwide processes. We cannot make the header info clusterwide as we dont have a mechanism right now to read the proc of a node remotely. Only means would be rexec or something like that. And displaying data of all node statistics in the header section of "top" would clutter the "top" window. Neither does keeping only the local data in the header and displaying clusterwide processes sound any good. We are thinking of ripping off the "top" header of all such fields that dont make good sense for remote processes. Any other suggestions are most welcome. "top" uses the "local" /proc/meminfo file to calculate %MEM for the remote processes, which is wrong. We are not targeting at fixing this right away. Am not sure, but we may hide the %MEM column in the output. 2) The new top will support --shownode option that would display a "NODE" column showing the execution node of a process. 3) top when run in localview mode (localview top) would show a normal top output with the headers and everything else, except that it will not show remote processes. Has anyone been seeing any problems with "top" on the ssi cluster ? Any suggestions on the above are most welcome. And Also if any of the things in the new implementation mentioned above would cause any problems. Thanks, Roopa |