Memory Page Faults

[Jim Rogers_5]

I am new to administration on a Unix 9000 server. This may be a dumb question, but I believe I am having a page fault problem. We commonly get comments that the server is operating slow. I am trying to identify the problem. One area I have seen is the page faults. below is a copy of the Glance Memory screen. Please, any help in identifying this issue will be greatly appreciated.

B3692A GlancePlus C.02.15.00 09:04:21 bmchp 9000/871 Current Avg High
--------------------------------------------------------------------------------
CPU Util S SU U | 87% 59% 100%
Disk Util F F | 12% 35% 100%
Mem Util S SU UB B | 64% 61% 65%
Swap Util U UR R | 61% 57% 89%
--------------------------------------------------------------------------------
MEMORY REPORT Users= 41
Event Current Cumulative Current Rate Cum Rate High Rate
--------------------------------------------------------------------------------
Page Faults 1497 964593 332.6 194.5 1983.6
Paging Requests 820 393591 182.2 79.3 780.3
KB Paged In 0kb 0kb 0.0 0.0 0.0
KB Paged Out 0kb 0kb 0.0 0.0 0.0
Reactivations 0 0 0.0 0.0 0.0
Deactivations 0 0 0.0 0.0 0.0
KB Reactivated 0kb 0kb 0.0 0.0 0.0
KB Deactivated 0kb 0kb 0.0 0.0 0.0
VM Reads 0 0 0.0 0.0 0.0
VM Writes 0 0 0.0 0.0 0.0

Total VM : 496.7mb Sys Mem : 151.8mb User Mem: 681.7mb Phys Mem: 1.38gb
Active VM: 284.8mb Buf Cache: 70.4mb Free Mem: 504.1mb
Page 1 of 1

[Pedro Cirne]

Hi,

I'm almost sure you don't have a swap problem.

You don't have any page out.

Please post the output of the "top" command.

Enjoy :)

Pedro

[Jim Rogers_5]

Here is the output from top.

System: bmchp Thu Aug 18 09:12:04 2005
Load averages: 1.32, 0.99, 0.79
343 processes: 340 sleeping, 3 running
Cpu states:
CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0 1.40 45.5% 0.0% 11.9% 42.6% 0.0% 0.0% 0.0% 0.0%
1 1.23 31.7% 0.0% 7.9% 60.4% 0.0% 0.0% 0.0% 0.0%
--- ---- ----- ----- ----- ----- ----- ----- ----- -----
avg 1.32 38.6% 0.0% 9.9% 51.5% 0.0% 0.0% 0.0% 0.0%

Memory: 502520K (286432K) real, 511296K (291040K) virtual, 520744K free Page# 1
/32

CPU TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
1 rroot 28400 oracle8 232 20 17524K 640K run 0:46 35.33 35.27 oraclePRO
1 rroot 28318 oracle8 217 20 17540K 660K run 0:43 27.38 27.34 oraclePRO
1 pty/ttyr4 22070 bheller 156 20 8808K 3608K sleep 0:17 13.69 13.67 quick
1 rroot 9339 oracle8 154 20 17572K 708K sleep 24:53 13.51 13.49 oraclePRO
1 rroot 22072 bheller 154 20 17660K 796K sleep 0:06 5.28 5.27 oraclePRO
0 rroot 1384 oracle8 154 20 245M 239M sleep 297:03 3.74 3.73 dbsnmp
1 rroot 77 root 100 20 0K 0K sleep 81:34 2.33 2.32 netisr
0 pty/ttyra 28611 bfairbro 156 20 7584K 2112K sleep 0:01 1.99 1.96 quick
1 pty/ttyp4 11596 wdyck 156 20 8156K 2944K sleep 0:14 1.47 1.46 quick
0 rroot 76 root 100 20 0K 0K sleep 184:45 1.27 1.27 netisr

[baiju_3]

Pagefault value seems to be normal .Please look number of deactivatios , if that number is high then you do have some memory issue.

[Pedro Cirne]

Hi,

You CPU load was not very high when you did the "top". CPU is mainly used by oracle processes, have you talked with you DBA?

Post the output of:

#sar -u 5 10

Enjoy :)

Pedro

P.S:Please don't forget to assign me point .-)

[Bill Hassell]

Memory page faults (from Glance) are not a problem in the quantity you're seeing. A page fault has nothing to do with RAM, it is an addressing mechanism. The default memory page size is 4k and each page could anywhere in memory. A set of ultra-highspeed registers are loaded with the addresses, but there are only a few dozen of them. For a really large program (dozens of megabytes), there aren't enough registers to cover the entire program space, so a page fault occurs when the program needs to access an area without a TLB entry. This means a reload of page register which takes nanoseconds, so there isn't a noticeable impact on performance. But if the program is a massively large and during execution, bounces randomly all over the address space, the page fault rate will be very high and can slightly slow down the execution speed, perhaps 2-9%. That's not a lot but for a numerical analysis program, it could mean a slower completion speed.

In your top listing, the system is running quite normally, not very much user CPU and a normal amount of system CPU, consistent with a database system. My guess is that your disks are very busy, again quite normal for a database. But if the users are complaining, it's time for the database admins to look at reducing disk I/O with more indexes, balancing indexes, and increasing the amount of RAM to buffer disk I/O. If you stop the database, the system probably runs at full speed.

[Jim Rogers_5]

This system only runs our Oracle DB. I will look into DB admin and disk I/O. Here is the output from sar you requested. The system is pretty free right now. We get hit in peaks. Then the CPUs are maxed out.

# sar -u 5 10

HP-UX bmchp B.10.20 A 9000/871 08/18/05

10:01:26 %usr %sys %wio %idle
10:01:31 45 35 3 17
10:01:36 36 22 3 39
10:01:41 15 3 2 79
10:01:46 16 3 1 80
10:01:51 25 9 1 65
10:01:56 32 16 3 49
10:02:01 24 39 4 33
10:02:06 9 5 1 85
10:02:11 16 3 2 79
10:02:16 14 3 1 82

Average 23 14 2 61
#