May it be that you took the average wait times from the “Avg wait (ms)” column of the “Wait events” table just below the “Top 5 timed events” section? There they are 1 and 0 – the down-rounded of the less than 2 and 1 millisecond you talked about?
Anyway, if this “Avg wait(ms)” column from the “Wait events” table means what I think it means – it is strange how they came up to this figures – as the number of “Waits” and “Total Wait Time” is the same as in the “Top 5 timed events” section – and by doing the arithmetic – the results are around 0.5 millisecond.

What do you think?

Sometimes you just have to do the arithmetic because the answer isn’t in the report – and then sometimes you do the arithmetic wrong. Look at the Top 5:

db file scattered read                            882,872         472    12.58
db file sequential read                           762,365         307     8.19

882,872 reads in 472 seconds (or 472,000 milliseconds) is 0.535 milliseconds per read, and 762,365 reads in 307,000 milliseconds is 0.40 milliseconds per read. I don’t know how I managed to get these wrong when I wrote the note. (Both my statements are true, of course, but not accurate – and I think I must have done one of the calculations upside down to get the 2 millisecond figure.

Your other two comments are really about use of language – I said that the direct path reads were “worst”, this doesn’t mean that they were absolutely bad. But as an “absolute” guideline – a physical read request that really goes to disc is behaving reasonably if it takes about six milliseconds to complete. When I said “I/O profile” it’s not intended as a highly technical expression – I’m just identifying the fact that there’s a pattern or anomaly that probably has some meaning; in this case the pattern is the very fast I/O response time that indicates the use of local filesystem cache.

Thanks for asking the question – it highlighted an error that I shouldn’t have made.

I have few doubts and queries..hope you will find some time to clarify those?

–> The I/O times are very informative. We are seeing lots of waits for single and multiblock reads, and some for direct path reads (which are often quite large).
But checking the average wait times, the single block reads are averaging less than one millisecond, and the multiblock reads are averaging less than 2 milliseconds.

Where do you see average wait times and the single block reads are averaging less than one millisecond, and the multiblock reads are averaging less than 2 milliseconds.?

–>The direct path reads are the worst at an average of 5 milliseconds

how should we conclude that average of 5 milliseconds for direct path reads is bad??

–> when you have an I/O profile like this
How do we say that this is I/O profile??

Apologies..this questions may sound naive..I have started perf tuning a year ago and just learning stuffs!!

your suggestions will be helpful!

Thanks,
Kartik

Physical reads = number of blocks read.
v$filestat.phyrds = number of read requests.

So it’s perfectly reasonable to see the first larger than the second.

Direct path reads can be asynchronous (and therefore not always reported with waits), so I would not have been suprised to see your read waits smaller than your reads; but yours are the other way round.

I can’t think of any obvious “real” reason for the difference, but these things are updated from different pieces of code, so I’m never too suprised to see small differences – especially since the dynamic performance views do not give you point-in-time read-consistency.

how simple looks when explained by you. Where are these physical reads comming from ? I have identified:
– datafiles reads( system,data,idx, undo…) ( most of phyrds)
– temp reads
– controlfile reads – few
– redologs reads – few
I would expect also for every read to have corresponding wait- so reads close to readwaits.
Next is a delta from 2 snapshots:

select name,value from v$sysstat where name =’physical reads’
union all
select ‘READS’,(select sum(phyrds) from v$filestat)+(select sum(phyrds) from v$tempstat) from dual
union all
select ‘READWAITS’,sum(total_waits) from v$system_event
where event in (‘db file sequential read’,'db file scattered read’,'db file parallel read’,'direct path read’,'direct path read (lob) ‘)

physical reads 402769 – ???
READS 263594
READWAITS 263663

Do I forget something?
Thank you.

Vasile