Roger Cariah wrote:
Here's what I see:
NETTYPE ipcshm,1,30,CPU # Configure poll thread(s) for nettype
NETTYPE soctcp,5,25,CPU # Configure poll thread(s) for nettype
You should never use CPU VPs for TCP listeners (ignore the manual).
Also shared memory listeners should reside in all CPU VPs, so, make this:
NETTYPE ipcshm,6,30,CPU # Configure poll thread(s) for nettype
NETTYPE soctcp,2,25,NET # Configure poll thread(s) for nettype
NUMAIOVPS 2 # Number of IO vps
This is too low, even with KAIO enabled. You will experience slowdowns
when lots of updates to the message log have to happen (ie when you are
busiest). Make it:
NUMAIOVPS 6 # Number of IO vps
Then monitor onstat -g iov. If there are more than one AIO VP with an
io/wup column value less than 1.0 you can drop some. If there are NO
AIO VPs with io/wup less than 1.0 you need to add more. Add two at a time.
Beyond that I'd like to see output from:
onstat -D
onstat -d
onstat -P
onstat -g iov
onstat -g iof
onstat -g glo
Art S. Kagel
<SNIP>