Kind regards,
Uwe

I am expecting from you.

Regards,
Ram

There are ways we can estimate some figures – but they all have holes in them.

SQL*Net roundtrips to/from client – Since 8i the OCI library has been able to bundle calls such as “close cursor”, parse, execute and even fetch into a single round-trip (a detail I avoided mentioning in my earlier comments), so we have room for error there; however, to a reasonable level of approximation, the number of user calls will be roughly the same as the number of executes plus fetches when a lot of single fetches are going on – which means the round-trips statistic doesn’t really add value.

The drawback to the v$sqlstats (et. al.) statistics is that cursors come and go, values get reset, and so on. We might get reasonable figures from summing them, we might get totally misleading figures. Whatever we get, we still have to apply some thought to what it means and how valid it might be – an actual instance activity stat would be so much more helpful.

This example is a little odd, since there’s very little going on at all and we haven’t even been given the “Top N timed events” section. However, when tracking down performance problems the general rule is to go for the most expensive things first and, as I pointed out in the thread, we have 652 user calls per second compared to the 48 parse calls (of which virtually none were hard parses) so the fetches – as Sid implied in his comment – are the most significant thing we could pursue.

You are correct, of course, that we could look at the “SQL ordered by Parse Calls” section if we really felt the need to pursue the parse calls.

In recent versions of Oracle the stats about reads and writes have been extended to clarify what’s going on here.

I think you’ll find that the writes reported in tablespace and file I/O stats are the number of I/O requests while the “Physical Writes” recorded in the load profile are the numbers for blocks read and written. SInce you can do multiblock writes to temp (and other direct path writes) the numbers in the load profile can be much higher than the numbers in the Tablespace and File I/O stats.

This picks up very nicely the difficulty of identifying the true source of almost any of the available statistics.

In this case you probably need to look at the instance activity stats for “recursive calls”. A single user call could be an execute call to a pl/sql procedure that then does lots of parse, execute and fetch calls.

Then there are the FETCHES, EXECUTIONS, and ROWS_PROCESSED columns of V$SQLSTATS (and similar) – one could easily use them to build some rational ratios, I think :-)

The Executes may larger than user call also, here is one load profile from a 11gR2 db, when server result cache is enabled(92.4 user calls per sec compare to 1449 executes). not sure how to format the code:-)

                Per Second      Per Transaction
DB Time(s):            3.6            0.3     
DB CPU(s):             1.1            0.1     
Redo size:       247,762.1       22,120.8                
Logical reads:    63,438.7        5,664.0                 
Block changes:     1,239.5          110.7           
Physical reads:    5,443.1          486.0           
Physical writes:      31.6            2.8             
User calls:           92.4            8.3             
Parses:               62.4            5.6           
Hard parses:           3.4            0.3             
W/A MB processed:     16.4            1.5                     
Logons:                0.5            0.1           
Executes:          1,449.0          129.4           
Rollbacks:             0.0            0.0             
Transactions:         11.2

I’ve just another question about the load profile. I found reports where the Physical Writes per second in the load profile didn’t match the numbers in the region Tablespace IO Stats. If I summarize the writes on every tablespace I come to 47 writes/s but in the load profile there are 397.8 writes/s listed.
The number of reads/s mathes nearly. 107.7 from the sum of tablespace vs. 120.6 from the load profile.

Are there any other writes included in the load profile?

Many thanks!

Thomas