|
Simply put, CPU is swifter than I/O (though not as
much as some would have us believe!).
7-8 years ago, we benchmarked Oracle 7.1 PQ on
single-processor IBM RS6000 nodes and found that the "sweet spot" was DOP=4
(where "DOP" means "degree of parallelism"). In testing on those 60Mhz
PowerPC RISC processors, one CPU on 10000RPM SSA drives (using RAID1+0) would
drive four PQ slave processes best. Adding more or removing any tended to
reduce overall performance. One CPU can easily drive several PQ processes,
because the bottleneck there is I/O...
I suspect that todays multi-Ghz processors, if
supported by the right backplane, bus, and RAM, can do a little better...
:-)
----- Original Message -----
Sent: Thursday, February 13, 2003 8:39
AM
Subject: RE: Re: parallel index creation
again:in which case, can we use p
Steve,
I
believe what you say, but it seems counter-intuitive.
If
you only have one cpu, and you start two jobs, then it follows that the cpu
needs to split itself to do the work.
So,
what are we gaining? The CPU can only go so fast and do so much
work.
Tom Mercadante
Oracle Certified Professional
tom,
if the process is IO
bound (ie consumes little cpu) then you can achieve a lot.
thanks,
steve
|
| "Mercadante, Thomas F"
<[EMAIL PROTECTED]>
Sent by: [EMAIL PROTECTED]
02/12/2003 02:23 PM
Please respond to ORACLE-L
|
To: Multiple recipients of list ORACLE-L
<[EMAIL PROTECTED]>
cc:
Subject: RE: Re: parallel index
creation again:in which case, can we use
p |
If you only have one CPU, then is parallel either not supported, or
simply a
waste of time?
I actually thought it was not supported.
If you only have one CPU, what do
you expect to gain?
Tom
Mercadante
Oracle Certified Professional
-----Original
Message-----
Sent: Wednesday, February 12, 2003 12:54 PM
To: Multiple
recipients of list ORACLE-L
p
My experience shows that a
parallel degree of less than 4 is nearly always
slower than serial.
I would recommend tring parallel degree of 4.
-----Original
Message-----
Sent: Wednesday, February 12, 2003 10:59 AM
To: Multiple
recipients of list ORACLE-L
parallel with single cpu
env?
Michael Ivanov,
Hi, Thanks for your reply.
In fact, I builded the index several times like, and the
result is
persistent across difference test case:
So, I think buffer is not the cause of the
parallel
execution slower. But I really do not get other parameter to
tune:(
SQL> set term on timing
on echo on feedback on
SQL> alter session set sort_area_size =
100000000;
Session altered.
Elapsed: 00:00:00.01
SQL>
create index idx_serial on viewcount( SID_LIST) nologging
parallel
(degree 2) tablespace pqind;
Index
created.
Elapsed: 00:18:01.36
SQL> drop index
idx_serial;
Index dropped.
Elapsed: 00:00:00.16
SQL>
SQL> create index idx_serial on viewcount( SID_LIST) nologging
tablespace
pqind;
Index created.
Elapsed:
00:06:48.04
SQL> drop index idx_serial;
Index
dropped.
Elapsed: 00:00:00.06
SQL>
SQL> create
index idx_serial on viewcount( SID_LIST) nologging parallel
(degree 2)
tablespace pqind;
Index created.
Elapsed:
00:14:51.92
SQL> drop index idx_serial;
Index
dropped.
Elapsed: 00:00:00.13
SQL>
SQL> create
index idx_serial on viewcount( SID_LIST) nologging;
Index
created.
Elapsed: 00:06:26.23
SQL> drop index
idx_serial;
Index dropped.
Elapsed: 00:00:00.06
SQL>
SQL> create index idx_serial on viewcount( SID_LIST) nologging
parallel
(degree 2) tablespace pqind;
Index
created.
Elapsed: 00:14:44.58
SQL> drop index
idx_serial;
Index dropped.
Elapsed: 00:00:00.13
SQL>
SQL> create index idx_serial on viewcount( SID_LIST) nologging
tablespace
pqind;
Index created.
Elapsed:
00:06:49.09
SQL> drop index idx_serial;
Index
dropped.
Elapsed: 00:00:00.07
SQL>
SQL> create
index idx_serial on viewcount( SID_LIST) nologging parallel
(degree 2)
tablespace pqind;
Index created.
Elapsed:
00:14:46.79
SQL> drop index idx_serial;
Index
dropped.
Elapsed: 00:00:00.14
SQL>
SQL> create
index idx_serial on viewcount( SID_LIST) nologging
tablespace
pqind;
Index created.
Elapsed:
00:06:44.51
SQL> drop index idx_serial;
Index
dropped.
Elapsed:
00:00:00.07
Regards
zhu
chao
msn:[EMAIL PROTECTED]
www.happyit.net
www.cnoug.org(China
Oracle User Group)
======= 2003-02-12 18:40:00 ,you
wrote£º=======
>Dear Chao.
>Did you try change order of
index's creating- first noparallel, second with
parallel. I think you
will look other results.
>
>> hi, dba friends:
>>
some paper
said, pqo should only be used in SMP machines, while
others
>>
say, We can also use pqo in uniprocessor machines in some case. I
am
trying
>> to use parallel index creation in the following
env:
>>
>> Dell 1650 with 3 scsi160 disks and 1 CPU and 2G
memory.
>> Oracle 9.2
>> Table contains 22000000
records,1.2GB
>> Table tablespace contains 3 datafiles , 400M, 400M
and 600M, on seperate
3
>> disks. Index tablespace contains 3
datafiles, 200M, 200M and 200M on
>> seperate 3
disks.
>>
>>
>> SQL> set term on timing on
echo on feedback on
>> SQL> alter session set sort_area_size =
100000000;
>> Session altered.
>> Elapsed:
00:00:00.01
>> SQL> create index idx_serial on viewcount(
SID_LIST) nologging parallel
>> (degree 2) tablespace pqind; Index
created.
>> Elapsed:
00:18:01.36
>> SQL> drop index idx_serial;
>>
Elapsed: 00:00:00.16
>> SQL> create index idx_serial on
viewcount( SID_LIST) nologging
tablespace
>> pqind; Elapsed:
00:06:48.04
>>
This machine is exclusived used my me and It seems that PQO
is
rather
>> slower than single thread. So is it still possible
to use PQO on single
>> processor machines? Please share your
experience and idear.
>>
Thanks.
>>
>> Wait event
like:
>>
>> Top 5 Timed Events
>>
~~~~~~~~~~~~~~~~~~
>> Total Event
Waits Time
(s)
>> Ela Time
-------------------------------------------- ------------
>>
----------- -------- PX qref latch
>> 48,371 415
40.94 PX Deq: Execute Reply
>> 176
340 33.54 PX Deq Credit: send
blkd
>> 47,704
248 24.47 control file parallel write
>>
112 5
.48 PX Deq Credit: need buffer
>>
1,835
4 .38
>>
-------------------------------------------------------------
^LWait
Events
>> for DB: ORA9 Instance: ora9 Snaps:
19 -20
>> -> s - second
>> -> cs - centisecond
- 100th of a second
>> -> ms - millisecond -
1000th of a second
>>
>>
>--
>Best regards
>Michael Ivanov, TD "ERA"
= = = = = = = =
= = = = = = = = = = = =
--
Please see the official ORACLE-L FAQ:
http://www.orafaq.net
--
Author: chao_ping
INET:
[EMAIL PROTECTED]
Fat City Network Services --
858-538-5051 http://www.fatcity.com
San Diego, California
-- Mailing list and web hosting
services
---------------------------------------------------------------------
To
REMOVE yourself from this mailing list, send an E-Mail message
to:
[EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the
message BODY, include a line containing: UNSUB ORACLE-L
(or the name of
mailing list you want to be removed from). You may
also send the
HELP command for other information (like subscribing).
--
Please see
the official ORACLE-L FAQ: http://www.orafaq.net
--
Author: Toepke,
Kevin M
INET: [EMAIL PROTECTED]
Fat City Network
Services -- 858-538-5051 http://www.fatcity.com
San Diego,
California -- Mailing list and web hosting
services
---------------------------------------------------------------------
To
REMOVE yourself from this mailing list, send an E-Mail message
to:
[EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the
message BODY, include a line containing: UNSUB ORACLE-L
(or the name of
mailing list you want to be removed from). You may
also send the
HELP command for other information (like subscribing).
--
Please see
the official ORACLE-L FAQ: http://www.orafaq.net
--
Author:
Mercadante, Thomas F
INET: [EMAIL PROTECTED]
Fat City
Network Services -- 858-538-5051 http://www.fatcity.com
San
Diego, California -- Mailing list and web hosting
services
---------------------------------------------------------------------
To
REMOVE yourself from this mailing list, send an E-Mail message
to:
[EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the
message BODY, include a line containing: UNSUB ORACLE-L
(or the name of
mailing list you want to be removed from). You may
also send the
HELP command for other information (like
subscribing).
|
