Yes. So notably my "ANDs" below are not in any way "BUTs."
AND, have you checked whether you have forced parallel local, or is the ETL
load itself putting load on the interconnect?
AND, if you are using connections to multiple nodes from various clients
executing the ETL jobs, are they operating on disjoint tables?
It is possible to minimize the "RAC" tax, the most likely contributors to
which JL has just documented. The notion of "Application Affinity" extends
to ETL: to the extent possible, do all the updates to the tables belonging
to application X on one node, Y on another node, and so forth, balancing the
load on nodes. Or if a single node will handles all the update load, just do
all the insert/update/deletes on a single node, so at least the gcc traffic
driven by queries from other nodes is at least unidirectional.
A little data model analysis and listener service configuration later, and
you can minimize your ractax.
Good luck!
mwf
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Jonathan Lewis
Sent: Tuesday, August 06, 2019 5:45 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Re: RAC - Global Cache Transfer Times
Definitely on the slow side.
The simplest rule of thumb is "if it's slower than a single block read we
probably shouldn't be using RAC".
You can see this type of repsonse (slow busy CR transfers) if you've got a
reporting node and an update node and the update node gets very busy. The
reporting node starts a "long" query and the update node has to do a lot of
work on pinning busy blocks, copying them, then making the copy
read-consistent.
In a single instance I think you'd see a lot of buffer busy waits, and a lot
of "consistent gets - undo records applied" (both of which you may see on
node 2 while node 1 is reporting these slow gc cr gets). Are you also
seeing a lot of "gc cr disk read" waits ?
Regards
Jonathan Lewis
________________________________________
From: oracle-l-bounce@xxxxxxxxxxxxx <oracle-l-bounce@xxxxxxxxxxxxx> on
behalf of Jack van Zanen <jack@xxxxxxxxxxxx>
Sent: 06 August 2019 01:08
To: oracle-l@xxxxxxxxxxxxx
Subject: RAC - Global Cache Transfer Times
Hi All,
Below is from AWR report and I was wondering if the CR Avg Time was on the
slow side.
Anyone have any rule of thumb as to what it "should" be?
BTW, this is during the ETL load into our DWH.
Global Cache Transfer Times (us)
* Avg Time - average time of all blocks (Immed,Busy,Congst) in us
* Immed, Busy, Congst - Average times in us
* ordered by CR + Current Blocks Received desc
CR Avg Time (us) Current Avg Time (us)
Inst No Block Class All Immed Busy Congst All Immed Busy
Congst
2 data block 26907 592 151759 942 392 9890
6391
2 undo header 2612 193 241481 1361 1361
2 others 11129 268 489014 54862 247 81818
2 undo block 201 201
Jack van Zanen
[https://docs.google.com/uc?id=0BwovDucFT1fXaEREVHNWRWZyNjg&export=download]
-------------------------
This e-mail and any attachments may contain confidential material for the
sole use of the intended recipient. If you are not the intended recipient,
please be aware that any disclosure, copying, distribution or use of this
e-mail or any attachment is prohibited. If you have received this e-mail in
error, please contact the sender and delete all copies.
Thank you for your cooperation
--
//www.freelists.org/webpage/oracle-l
--
//www.freelists.org/webpage/oracle-l
