[Search for users]
[Overall Top Noters]
[List of all Conferences]
[Download this site]
| Title: | POLYCENTER Performance Advisor for Digital UNIX |
|
| Moderator: | PECSYS::BHAT |
|
| Created: | Fri Feb 17 1995 |
| Last Modified: | Mon Jun 02 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 262 |
| Total number of notes: | 856 |
259.0. "hard page faults" by ISTWI1::ASLANTEKIN () Wed Apr 30 1997 12:45
Hi,
One of our customers has performance problems on AlphaServer 8400.When we watch
system performance with Polycenter Performance Advisor, we get the following
analysis report.
Configuration:
2 440 Mhz.Dual CPU
2 GB memory
3.5 GB swap space
Digital Unix V3.2D-1
Many KZPSA,HSZ40 and RZ29 disks
We would like to be sure we really have to add more memory? What is the
meaning of hard page faults and apmle free memory? How can decrease number of
hard page faults? The customer is a GSM company and The performance of the
system is really important for us.
Thanks for any help,
Umit
Full Analysis turkcell Page 1
PSPA V3.0-7
Tuesday 29-Apr-1997 00:00 to 29-Apr-1997 14:14
CONCLUSION 1.
{M0060}
The large number of hard page faults occurring may indicate
a memory shortage problem.
If your capacity plan indicates a growth in demand, you may
want to consider adding memory to the system. If ample free
memory exists, find out why secondary cache is used instead
of primary cache to satisfy the process's demands for pages.
Total number of occurrences: 14
CONDITIONS
1. VPF_PAGEFAULTS .GE. ( 3.000000 * HW_SPEED ) .OR.
VPF_HARDFAULTS .GE. 13.000000
2. VPF_HARDFAULTS .LT. 13.000000 .OR. VPF_FREEPAGES
.GT. VM_PAGE_FREE_MIN .OR. ( RSS (
TOP_PROCESS_RSS_X ) .LT. VPF_TOTALPAGES * 0.220000
) .AND. VPF_PGWRITES .LT. 13.000000 / 3.000000
3. IMAGE_ACTIVATION_RATE .LT. 20.000000
4. VPF_PAGEFAULTS .LT. ( 3.000000 * HW_SPEED )
5. VPF_HARDFAULTS .GE. 13.000000
6. OCCURRENCES .GE. 5
EVIDENCE
Pagfile Page Free UBC PagFile PagFile Pages Pages Time
IO Rate Faults Pages Pages WritIO ReadIO Writn Read
------- ------ ----- ----- ------- ------ ----- ----- ---------------
32.767 72.875 22074 176916 0.000 32.767 0.000 33.68 29-Apr 02:08:00
88.408 121.90 8183 179821 0.000 88.408 0.000 89.03 29-Apr 02:56:00
48.758 85.825 17915 179821 0.000 48.758 0.000 49.65 29-Apr 02:58:00
16.388 30.621 22605 179821 0.000 16.388 0.000 17.06 29-Apr 05:40:00
33.475 183.51 11591 166382 0.000 33.475 0.000 44.69 29-Apr 10:10:00
103.800 294.02 26621 171871 0.000 103.80 0.000 116.3 29-Apr 10:12:00
19.525 503.73 64 173695 4.767 14.758 1.525 28.14 29-Apr 13:52:00
76.833 462.41 13734 179810 0.000 76.833 0.000 88.11 29-Apr 14:04:00
50.674 414.70 3693 179821 0.815 49.859 0.289 62.53 29-Apr 14:06:00
52.171 508.14 60 179000 2.438 49.733 0.781 63.25 29-Apr 14:08:00
85.633 517.55 13181 179821 0.000 85.633 0.000 102.7 29-Apr 14:10:00
51.800 431.00 2426 179821 0.375 51.425 0.108 65.10 29-Apr 14:12:00
32.558 467.98 58 178654 0.000 32.558 0.000 46.71 29-Apr 14:14:00
88.283 541.68 12828 179821 0.000 88.283 0.000 103.6 29-Apr 14:16:00
CONCLUSION 3.
{C0030}
A CPU bottleneck occurred when the CPU time to perform
memory management (service pagefaults and swap processes)
became excessive.
To alleviate the bottleneck, either:
o Examine memory usage and application design to reduce
memory demand.
o Reschedule the workload to avoid memory contention.
o Add memory to the system.
Total number of occurrences: 12
CONDITIONS
1. THR_RUNABLE .GE. 6.000000
2. CPU_PERCENT ( TOP_PROCESS_CPU_TIME_X ) .LT. 40.000000
3. CPU_TOTINTS .LT. 2000.000000
4. CPU_SYSTIME .GE. 25.000000
5. VPF_PAGEFAULTS .GE. ( 3.000000 * HW_SPEED ) .OR.
VPF_HARDFAULTS .GE. 13.000000
6. OCCURRENCES .GE. 5
EVIDENCE
Pg Pg
CPU CPU CPU CPU PgFil File File
CPU SYS User Nice Idle I/O Page Free UBC Writ Read
Que. Time Time Time Time Rate Flts Pages Pages I/O I/O Time
---- ---- ---- ---- ---- ----- ---- ----- ----- ---- ---- ---------------
8.862 26.5 6.53 13.4 53.6 88.41 122 8183 ***** 0.00 88.4 29-Apr 02:56:00
11.17 25.4 0.20 2.31 72.1 16.39 30.6 22605 ***** 0.00 16.4 29-Apr 05:40:00
7.931 26.0 15.8 0.56 57.5 33.47 184 11591 ***** 0.00 33.5 29-Apr 10:10:00
8.379 26.3 17.9 9.77 46.0 103.8 294 26621 ***** 0.00 104 29-Apr 10:12:00
12.39 34.1 23.9 7.84 34.2 19.52 504 64 ***** 4.77 14.8 29-Apr 13:52:00
12.62 30.5 29.9 17.5 22.1 76.83 462 13734 ***** 0.00 76.8 29-Apr 14:04:00
10.39 28.2 26.4 14.5 30.9 50.67 415 3693 ***** 0.81 49.9 29-Apr 14:06:00
9.400 26.3 26.1 16.9 30.7 52.17 508 60 ***** 2.44 49.7 29-Apr 14:08:00
9.552 28.0 26.6 14.9 30.6 85.63 518 13181 ***** 0.00 85.6 29-Apr 14:10:00
9.310 27.4 25.6 19.3 27.7 51.80 431 2426 ***** 0.38 51.4 29-Apr 14:12:00
11.66 33.6 31.3 18.2 16.9 32.56 468 58 ***** 0.00 32.6 29-Apr 14:14:00
9.345 25.9 23.8 21.2 29.2 88.28 542 12828 ***** 0.00 88.3 29-Apr 14:16:00
Full Analysis turkcell Page 4
PSPA V3.0-7
Tuesday 29-Apr-1997 00:00 to 29-Apr-1997 14:14
CONCLUSION 4.
{R0080}
The user who used the most number of processes has been consuming
most of the CPU time. This kind of unbalanced process load is
affecting the workload of the other users on the system.
To alleviate the problem, either:
o Reschedule CPU intensive works for the user to avoid unbalanced
process workload among the other users.
o Adjust the scheduling priority of less important jobs.
o Change the system parameter autonice to ease the contention among
users for CPU time.
Total number of occurrences: 1
CONDITIONS
1. THR_RUNABLE .GE. 6.000000
2. NUM_USERS .GT. 2.000000
3. USER_NAME ( TOP_PROCESS_CPU_TIME_X ) .EQS.
USER_NAME ( TOP_PROCESSES_USER_X )
4. CPU_PRCNT_FOR_TOP_USER .GE. 80.000000
5. OCCURRENCES .GE. 1
EVIDENCE
Top User
----------------------------- Total Number Total Number
User Name %CPU # of Procs of Users of Procs Time
--------- -------- ---------- ------------- ------------ ---------------
root 89.461 58 5 115 29-Apr 06:34:00
| T.R | Title | User | Personal
Name | Date | Lines
|
|---|