I'm using a new Solaris machine using the Niagara CPU. bcp seems to be
running much slower then other platforms. *much* slower. A bcp on the
Niagara platform seems to take twice as long as an older Sparc CPU.

I'm using Sybase 12.5 and Solaris 10.




sun4u

syscall seconds calls errors

_exit .000 13

read .008 286

write 4.945 50485

open .295 88 10

close .003 148 7

time .000 6

chmod .000 2 2

brk .002 90

stat .002 53 4

lseek .000 3

getpid .000 39

getuid .000 5

access .000 12 2

pipe .000 7

getgid .000 2

ioctl .000 12 9

execve .011 9

umask .000 2

fcntl .000 49

fcntl .000 11

fcntl .000 6

readlink .000 3 3

sigaction .001 79

sigfillset .000 2

getcontext .000 12

setcontext .000 6

setustack .001 9

waitid .000 20 5

mmap .004 164

munmap .001 64

setrlimit .000 1

getrlimit .000 13

memcntl .002 30

sysinfo .000 1

fork1 .004 12

lwp_self .000 12

lwp_sigmask .001 70

llseek .000 3 2

pollsys .000 5

door_info .000 1

door_call .000 1

schedctl .000 8

resolvepath .001 58

stat64 .001 26 8

lstat64 .000 2

fstat64 .000 20

open64 .001 7

so_socket .000 1

connect .000 2 2

recv 2.740 52580 1

send .000 6

setsockopt .000 5

-------- ------ ----

sys totals: 8.040 104551 55

usr time: 101.964

elapsed: 148.950

bash-3.2$



And the same script running on globes1, a sun4v system:



bash-3.2$ truss -fc ksh ./test_bcp.sh

************************************************** ********************

start: Thu Sep 11 16:19:10 GMT 2008

end: Thu Sep 11 16:23:01 GMT 2008

************************************************** ********************

signals ------------

SIGCLD 4

total: 4





syscall seconds calls errors

_exit .000 13

read .011 274

write 1.516 3169

open .029 87 11

close .003 151 7

time .000 6

chmod .000 2

brk .002 98

stat .001 49 4

lseek .000 1

getpid .000 40

getuid .000 5

access .000 10 1

pipe .000 7

getgid .000 2

ioctl .000 13 9

execve .013 9

umask .000 2

fcntl .000 51

fcntl .000 11

fcntl .000 6

readlink .000 4 4

sigaction .001 79

sigfillset .000 2

getcontext .000 12

setcontext .000 7

setustack .001 9

waitid .001 20 4

pathconf .000 2

mmap .005 146

munmap .005 54

setrlimit .000 1

getrlimit .000 13

memcntl .004 25

sysconfig .000 1

sysinfo .000 2

fork1 .006 12

lwp_self .000 12

lwp_sigmask .001 71

llseek .000 3 2

pollsys .000 6

schedctl .000 8

resolvepath .002 53

stat64 .001 28 8

lstat64 .000 2

fstat64 .000 19

open64 .000 7 1

so_socket .000 3

connect .000 3 2

recv 2.862 38419 1

recvfrom .000 1

send .000 7

setsockopt .000 5

-------- ------ ----

sys totals: 4.479 43042 54

usr time: 218.050

elapsed: 230.660

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.

Hello,

Are you running bcp out or in?

One platform looks like it is doing many more write() calls than the
other:
sun4u
syscall seconds calls errors
write 4.945 50485
recv 2.740 52580 1

And the same script running on globes1, a sun4v system:
syscall seconds calls errors
write 1.516 3169
recv 2.862 38419 1

Rather than use "truss -c", if you look at a detailed truss, can you
see
if they are writing the same amount of data with each call? For
example,
if you were doing bcp out, find the file you are writing to and look
to see
how many bytes are being written out:
27030: open("aut.dat", O_WRONLY|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) =
4
27030: write(4, " 1 7 2 - 3 2 - 1 1 7 6\t".., 1833) = 1833

I would also be curious about the recv() calls. One of your sessions
does way more than the other. So it would look like either they are
using different packet sizes or there are different amounts of data
being read.