VirtualBox

Changes between Initial Version and Version 1 of Ticket #6990


Ignore:
Timestamp:
06/14/10 11:06:37 (4 years ago)
Author:
frank
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #6990 – Description

    initial v1  
    11There's nothing much available in terms of AIOMgr documentation. Some of the things I'd like: 
    22 
    3 1) recommended system settings to enable AIO (e.g. kern.aioprocmac sysctl on OS X) 
    4 2) several settings are referenced in the logs (cache commit interval, cache commit threshold), but it's not clear where these settings are made and where it might be appropriate to change them 
    5 3) it's not clear where to get metrics to determine whether tunings are working as expected or where problems may crop up in the I/O path 
     3 1. recommended system settings to enable AIO (e.g. kern.aioprocmac sysctl on OS X) 
     4 2. several settings are referenced in the logs (cache commit interval, cache commit threshold), but it's not clear where these settings are made and where it might be appropriate to change them 
     5 3. it's not clear where to get metrics to determine whether tunings are working as expected or where problems may crop up in the I/O path 
    66 
    77In terms of metrics and debug, I've got a number of cases where CPU utilisation on Windows guests maxes out in the middle of disk I/O, and I can't really tell at what level the problem occurs. I noticed that SAS controllers don't produce metrics at all. Perhaps more importantly, the primary metrics interfaces don't appear to show disk I/O (I can only find this through the session information dialog in the GUI) or allow me to understand how the queuing looks to the guest or how it's being passed on to the host. I can't really understand whether there's some synchronisation problem in the drivers in the guest that are causing the problem or whether the guest is queuing I/O faster than the host is dispatching it, leading Windows to throw a tantrum because I/Os are appearing to time out or some such. As this is an issue where the guest hangs without giving me much in the way of data about whatever it was that caused it to hang (and this is in the middle of things like running Windows Update, so it's not the kind of case where a hang would be expected), it would be extremely useful to be able to observe the guest behaviour as it interacts with the virtualisation layer to get some idea of what's going on in there. 

www.oracle.com
ContactPrivacy policyTerms of Use