%busy in sar -d is high.

[chuanpeng.wang]

The %busy is high on disk c8t1d2 and the application is slow.
The application is informix, using raw file.
The IO on disk seems not heavy.
Is it a problem?

# sar -d 2 40

HP-UX data B.11.11 U 9000/800 01/30/08

15:17:45 device %busy avque r+w/s blks/s avwait avserv
Average c8t0d2 13.02 0.50 12 47 0.00 11.83
Average c8t0d4 13.68 0.50 12 48 0.00 12.13
Average c8t0d6 12.27 0.50 12 49 0.00 10.79
Average c8t1d0 12.47 0.50 12 48 0.00 11.01
Average c8t1d2 86.94 0.50 193 773 0.00 8.85
Average c8t1d4 6.79 0.50 5 21 0.00 12.96
Average c8t1d6 7.96 0.50 6 25 0.00 13.47
Average c8t2d0 7.19 0.50 6 22 0.00 13.30
Average c8t2d2 7.95 0.50 6 23 0.00 14.33
Average c8t2d4 65.42 0.50 74 296 0.00 13.42
Average c8t2d6 4.39 0.50 5 191 0.00 9.23
Average c8t3d0 9.90 0.59 16 113 0.16 7.04
Average c9t0d1 14.10 0.50 12 51 0.00 12.12
Average c9t0d3 12.25 0.50 11 45 0.00 11.55
Average c9t0d5 12.72 0.50 12 50 0.00 10.99
Average c9t0d7 12.18 0.50 12 49 0.00 10.51
Average c9t1d1 26.28 0.50 21 82 0.00 14.29
Average c9t1d3 7.52 0.50 6 23 0.00 13.18
Average c9t1d5 8.34 0.50 6 26 0.00 13.24
Average c9t1d7 8.09 0.50 6 24 0.00 13.72
Average c9t2d1 7.76 0.50 6 24 0.00 13.50
Average c9t2d3 43.04 0.50 36 147 0.00 14.88
Average c9t2d5 5.64 0.50 6 255 0.00 9.04
Average c2t1d0 1.65 1.57 4 127 1.67 7.39
Average c3t0d0 1.05 1.87 3 122 1.72 5.26
Average c9t2d7 0.07 0.50 0 0 0.00 18.40


sar -Mu 2 20

HP-UX data B.11.11 U 9000/800 01/30/08

15:23:43 cpu %usr %sys %wio %idle
Average 0 3 3 60 34
Average 1 5 27 39 29
Average 2 8 29 15 49
Average 3 5 3 22 70
Average 4 14 57 17 12
Average 5 17 55 16 12
Average 6 4 3 58 35
Average 7 3 4 59 35
Average system 7 22 36 35

[Steve Lewis]

blks/s divided by r+w/s is about 4, which tells me that it is doing lots of random access from that disk. Sequential access is made in reads of larger blocks per system call, defined by onconfig parameter RA_PAGES etc.

You don't tell us what that disk is in reality. Is it a LUN on a disk array?
Have you tried increasing the BUFFERS onconfig parameter?
Other things to look into:
1. Increase SAN cache to accommodate more data.
2. If it contains multiple tables which are frequently joined by SQL, then move one table elsewhere. If it contains data and logs, then move the logs elsewhere.
3. Try fragmenting (=Oracle partitioning) the table.
4. Try striping the data on this disk to more disks, using LVM or VXVM. That will reduce the number of operations to this disk and reduce the busy%.
5. There's lots more to look into, especially in the database (delayed logging, log striping, log buffers...etc) if the load is write-based.

I am guessing that the problem may be write-based to logical logs or the physical log due to usage of RAID 5 and slow service time for writes in this RAID configuration. No raid 5 for logs! Its the same for Oracle, Sybase, Ingres, Informix, MySQL etc... If it is a disk array what is the RAID config?