Re: What MB/s is pulled from your buffer cache?
From: Tanel Poder <tanel_at_tanelpoder.com>
Date: Thu, 14 Feb 2013 01:49:49 +0200
Message-ID: <CAMHX9JKmTfZRjpFDaCDH67WYA_fjjyQf3pT+E=Ac4sfENUVdtQ_at_mail.gmail.com>
By the way the "logical read bytes from cache" metric doesn't mean that this amount of bytes were actually processed / transferred from buffer cache memory to CPUs. An example would be a nested loop join or a filter loop, which probes some index/table millions of times in a loop - taking only one (or a couple of) rows each time. So, on each iteration you could only be touching only a couple of hundred bytes worth of memory lines in that 8kB buffer each time (btw I'm not talking about all the other memory access that's going on when running oracle code).
Date: Thu, 14 Feb 2013 01:49:49 +0200
Message-ID: <CAMHX9JKmTfZRjpFDaCDH67WYA_fjjyQf3pT+E=Ac4sfENUVdtQ_at_mail.gmail.com>
By the way the "logical read bytes from cache" metric doesn't mean that this amount of bytes were actually processed / transferred from buffer cache memory to CPUs. An example would be a nested loop join or a filter loop, which probes some index/table millions of times in a loop - taking only one (or a couple of) rows each time. So, on each iteration you could only be touching only a couple of hundred bytes worth of memory lines in that 8kB buffer each time (btw I'm not talking about all the other memory access that's going on when running oracle code).
This is different from the "physical read total bytes" metric in the sense that when you read 8kB from disk, you have to read it all in (and checksum it from end to end) even if you plan to read only one row from it.
-- *Tanel Poder* Enkitec (The Exadata Experts) Training <http://blog.tanelpoder.com/seminar/> | Troubleshooting<http://blog.tanelpoder.com/> | Exadata<http://www.amazon.com/Expert-Oracle-Exadata-Apress/dp/1430233923> | Voicee App <http://voic.ee/> On Wed, Feb 13, 2013 at 5:12 PM, Rich Jesse < rjoralist2_at_society.servebeer.com> wrote:Received on Thu Feb 14 2013 - 00:49:49 CET
> Martin writes:
>
> > So it's 4,970,209,998 Bytes per second overall. Fairly less than yours.
> :)
>
> At 4GB/s, that's roughly 40 times my number. :)
>
> My server uses DDR2 PC2-5300 RAM (man, would I love to upgrade), which
> gives
> it a theoretical max bandwidth of 10.6GB/s, assuming a dual-channel
> configuration. I'm not sure how an IBM p-series is architected CPU-to-RAM,
> but I just saw a peak transfer of over 13GB/s.
>
> Fun stuff. Thanks!
>
> Rich
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
-- http://www.freelists.org/webpage/oracle-l