Physical Writes is more than Physical Reads?

[public]

Hi,

I use perf. agent v4 to collect performance data on the Linux AS V2.1 server with Oracle 8.1.7. The system is never been tuned and I find the number of phy. writes is more than phy. reads, refer with attach file. Does anyone know why the phy. writes is more than phy. reads? How to make the phy. reads more than phy. writes?

Best Regards,
Eric

[Jerome Henry]

Hi,

Can be that Oracle needs to write changes and store read files in a stack (or in RAM), so reading the same datas quite often doesn't imply physicial read access to disk, but write does.
You can tune these parameters, if you feel it necessary, with the elvtune command :
/sbin/elvtune -r 1024 -w 2048 /dev/hda3
will require 1024 cycles latency between 2 reads, and 2048 cycles between 2 writes on hda3.
r should be about half write value to optimize head moves on the disk (storing writes on a stack).
You may then perform a bench moving these values to check if it improves disk speed.

hth

J

[Michael Schulte zur Sur · Accepted solution]

Hi,

the redologs are a busy place, where anything changed is logged including rollback segments. So it is very well possible to have more writes than reads.

Michael

[Michael Schulte zur Sur]

Hi,

one more thing: Be happy, when physical reads are few. Look for logical reads to see, how well tuned your instance is.

Michael

[public]

Michael,

Thanks your reply,I find the Oracle buffer cache hit ratio is negative:
Consistent_Gets DB_Blk_Gets Physical_Reads Hit_Ratio
4,907,767,584 252,769,373 5,789,148,480 -12.181

The phy._read is more than consistent_get that means Oracle doesn't use buffer cache!

What's something wrong?

Best Regards,
Eric

[Michael Schulte zur Sur]

Hi,

I have ran into your posting on metalink. ;-)
I have found something here, which you should read. Perhaps it is related to your case:
http://metalink.oracle.com/metalink/plsql/ml2_documents.showDocument?p_database_id=NOT&p_id=33883.1

greetings,

Michael

[public]

Michael,

Thanks your reply, I use the common method to get the hit_ratio:
select sum(decode(NAME, 'consistent gets',VALUE, 0)),
sum(decode(NAME, 'db block gets',VALUE, 0)),
sum(decode(NAME, 'physical reads',VALUE, 0)),
round((sum(decode(name, 'consistent gets',value, 0)) +
sum(decode(name, 'db block gets',value, 0)) -
sum(decode(name, 'physical reads',value, 0))) /
(sum(decode(name, 'consistent gets',value, 0)) +
sum(decode(name, 'db block gets',value, 0))) * 100,2)
from sys.v_$sysstat;


I only use "default" pool,no any keep or recycle pool buffer!

Best Regards,
Eric