On Sunday 13 August 2006 12:31, Richard Andrews wrote: > This is serverland. X doesn't exist. > ntpd maintains the clock. Yes but unless you don't tune it, the distro scripts will try (and fail) to run hwclock at bootup and shutdown.
> How does hdparms interact with UML? > Do UBD devices appears as a type of > disk? They are disks... but hdparm should fail - UML devices appear as block devices like disks, but they support less ioctl commands (just the one for partitioning support, i.e. geometry querying). And to my knowledge hdparm uses just ioctl() on disks. However test that - there was an unconfirmed report of a single case (probably solved since then) of Uml interacting with the host disk at shutdown - we've identified at that time a security issue which could have maybe caused that, if somebody tried to break in that UML. That said, hdparm -tT will work just fine on ubd disks (it works probably on any block device). > > Partition tables: you'd normally copy a single partition rather than a > > disk, so you don't get partition tables. > > The reason for this is that when you have a partition table on a file, > > mount -o loop does not parse the table, so you need a script (easily > > writable) to calculate the offset and pass it as offset=$n_sectors to > > mount or losetup. And this is not widely known. > What about multiple filesystem image files. Can I do: > ubd0=fs0.image ubd1=fs1.image ubd2=... Yes, it works, sorry for not mentioning that. That's what most of us does. > The server system is strictly segregated with /var, /usr, etc partitions > and disk usage constraints and potential overflow conditions are of > significant interest to me in terms of the simulation. > What about eg. databse IO tuning? Obviously time is measured in terms of > the guest kernel; is there anything useful to be gained from watching IO > behaviour in UML. The UBD_SYNC option (i.e. disabling write caching done by the host) is very costly, but you can enable it per-disk; it avoids data loss when the _host_ crashes. For journalized fs's that's theoretically important, but nobody reported problems with it disabled when using ext3 in UML; somebody reported problems with reiserfs in UML and it disabled. For the database transaction logs, however, you can't do without it. So for the database you should do something like: ubd5s=dbLogImage.fs (note the "s" after 5), put the logs on that partition, and possibly the dbLogImage.fs on another disk (if you have one); and probably for performance you do not want to create sparse images for heavily written ones - they'd cost time at first writes, but moreover they risk probably being more fragmented. > Typically this is more of an app specific thing though and I just check > ctime. Just looking for special conditions - I figure people on this list > would know the unwritten facts. Never discussed this issue, that's why. -- Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!". Paolo Giarrusso, aka Blaisorblade http://www.user-mode-linux.org/~blaisorblade Chiacchiera con i tuoi amici in tempo reale! http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ User-mode-linux-user mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
