[brandz-discuss] Oracle undex lx brand is very slow

[Edward Pilatowicz]

hey dmitry,
so i haven't run oracle under brandz in ages, and i don't remember what
version i ran when i did.  given that there's a supported version
of oracle that runs under solaris, i'd recommend running it in a
native zone.

but if you want to run the lx version, and you want to know what's
going on with the nanosleep, you should try getting a stack trace
from the nanaosleep call.  you can do this using truss/pstack
or dtrace.  with truss, you could use the -T option to stop the
process when it does a nanosleep call, get the stacktrace via
pstack, and then restart the process via prun.  or you could just
use dtrace and do something like:
	dtrace -n 'syscall::nanosleep:entry/execname == "rman"/{ustack(20)}'

ed

On Thu, Feb 05, 2009 at 09:34:25PM -0800, Dmitry wrote:
> Hi all,
>
> Does anybody have oracle working undex lx brand (centos from this site)?
>
> I have Solaris 10 5/08 and lx brand zone on it. I`ve install ORACE 10.2 inside and try to duplicate DB by rman. When I copy backup to linux I see over 30Mb/s write speed. Then I start rman to restore DB and it write over 1Mb/s. It`s very slow.
>
> I try to truss rman process in global zone and see:
>
> fstat64(16, 0x07FE4910)                         = 0
> read(16, "\0A3\0\006\0\0\0\0\00601".., 2064)    = 163
> uucopy(0x07FEF508, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00020000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> uucopy(0x07FEF238, 0x07FEF488, 8)               = 0
> uucopy(0xFEFFD36C, 0x07FEF28C, 20)              = 0
> uucopy(0x07FEF488, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00000000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> nanosleep(0x07FEF3F0, 0x07FEF3F0) (sleeping...)
> nanosleep(0x07FEF3F0, 0x07FEF3F0)               = 0
> write(16, "\0 '\0\006\0\0\0\0\003 N".., 39)     = 39
> fstat64(16, 0x07FE4910)                         = 0
> read(16, "\0A3\0\006\0\0\0\0\00601".., 2064)    = 163
> uucopy(0x07FEF508, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00020000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> uucopy(0x07FEF238, 0x07FEF488, 8)               = 0
> uucopy(0xFEFFD36C, 0x07FEF28C, 20)              = 0
> uucopy(0x07FEF488, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00000000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> nanosleep(0x07FEF3F0, 0x07FEF3F0) (sleeping...)
> nanosleep(0x07FEF3F0, 0x07FEF3F0)               = 0
> write(16, "\0 '\0\006\0\0\0\0\003 N".., 39)     = 39
> fstat64(16, 0x07FE4910)                         = 0
> read(16, "\0A3\0\006\0\0\0\0\00601".., 2064)    = 163
> uucopy(0x07FEF508, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00020000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> uucopy(0x07FEF238, 0x07FEF488, 8)               = 0
> uucopy(0xFEFFD36C, 0x07FEF28C, 20)              = 0
> uucopy(0x07FEF488, 0x07FEF238, 8)               = 0
> lwp_sigmask(SIG_SETMASK, 0x00000000, 0x00000000) = 0xFFBFFEFF [0x0000FFFF]
> nanosleep(0x07FEF3F0, 0x07FEF3F0) (sleeping...)
>
> Every nanosleep call it delay around 5 sec. Why?
>
> Can anybody help me to solve my problem with slow rman writing?
>
> I am ready to put other info about my configuration.
> --
> This message posted from opensolaris.org
> _______________________________________________
> brandz-discuss mailing list
> brandz-discuss at opensolaris.org