rzi@sposp1.UUCP (Roman Zielinski) (03/01/90)
We are trying to make some performance measurements on unix systems, but we had stated that the official V.3 environment is infantly primitive in this area - the only reasonable (but not the best is 'sar'). What I'm looking for, are tools allowing you to see: - total cpu load, momentary and integrated during the measurement period - cpu load per process and kernel, momentary and integrated - total disk load, momentary and integrated - disk load per partition/subvolume - average and max access times for disks, io-sizes etc - sizes of processes (text, data, bss ...other if you have a smart linker), momentary - 'fame' files per file system (ie names of files + number of accesses per measurement period) - file fragmentation - utilization of kernel buffer pools (occupation, fragmentation...) - currently used swapping strategy/policy I've earlier worked with performance on V.2-like systems from Philips (MPX P9X00) - MPX had all those fancy tools with histogram presentation etc so all system tuning and optimization was quite nice. But unfortunately those tools cannot be ported too easy to the 'real' unix. Please, all wonderful wizards, give us some good ideas (of cause sources are also very welcome!!!). Greetings from sunny Stockholm, where spring came at least 3 months to early (i.e. no skiing). Roman +---------------------------------+-------------------------------------------+ ! Roman M. Zielinski ! <here should be something wise...> ! ! Philips Tele & Data System AB ! <but there is not> ! ! S-115 84 Stockholm, Sweden ! ! ! tel +46 8 782 1373 ! ! ! uunet!mcsun!sunic!sposp1!rzi ! ! +---------------------------------+-------------------------------------------+
chrisl@fulcrum.bt.co.uk (Chris Parkin Lilley [cs12]) (03/02/90)
In article <251@sposp1.UUCP> rzi@sposp1.UUCP (Roman Zielinski) writes: >We are trying to make some performance measurements on unix systems, but we had >stated that the official V.3 environment is infantly primitive in this area - >the only reasonable (but not the best is 'sar'). > >What I'm looking for, are tools allowing you to see: [etc. etc. ] I would also be interested in any pointers to information on tuning (particularly with reference to V.3/680X0) cos' I'm going to have to do some real soon. Anticthanksipation -Chrisl. -- "Time is an illusion. Lunchtime doubly so." -------------------------------------------------------------------------------- Chris Parkin Lilley, CS1.2, BT Fulcrum, UK chrisl@uk.co.bt.fulcrum