+42 (Thanks Stephan, for keeping this clear.)
AND this benchmark is important for any actual configuration: Even if some
vendor expands the limits of a single bus to their SSD LUN beyond the speed
(latency and bandwidth) of SSD or any future persistent memory, it is still
very likely connected to something on the host computer that is much faster as
multiple separate channels from the persistent memory into your computer RAM.
The plumbing is complicated with many layers of competing latency and
bandwidth. This is a good reason to think good thoughts about Kevin Closson for
the end to end actual performance RESULT from the stack for Oracle via his SLOB.
BORING is good, be a SLOBber.
mwf
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On ;
Behalf Of Stefan Koehler
Sent: Thursday, June 25, 2020 6:30 AM
To: tgascard@xxxxxxxxx; ORACLE-L
Subject: Re: Lun size and performance with flash disks : rh7
Hello Thierry,
first counter question for storage vendor would be: Can you please provide a
benchmark (with raw data if possible) that proves your statement?
I mean it is always important to check what they are referring/comparing to.
Comparing one SSD/Flash LUN to some old rusty spindle disks ... well yes ... in
this case SSD/Flash might always be faster and you are happy anyway.
However if you compare one SSD LUN vs. ten SSD LVM striped LUNs with +15K IOPS
or so ... well then the picture is different.
This topic is pretty easy to verify. Just run your workload (e.g. SLOB
benchmark) on a file system backed by one big SSD LVM LUN and on file system
backed by ten striped LVM SSD LUNs and monitor with iostat -xk 1 <LUNs> ....
you gonna see the difference for sure.
Based on my experience with a lot of performance benchmarks - multiple striped
(SSD) LUNs make an impact on systems with big I/O load :-)
Best Regards
Stefan Koehler
Independent Oracle performance consultant and researcher
Website: http://www.soocs.de
Twitter: @OracleSK
thierry gascard <tgascard@xxxxxxxxx> hat am 25. Juni 2020 um 08:51--
geschrieben:
Hello ,Storage vendors claim that we do not care anymore about io queue per
lun on scsi.Two years ago I made test on ssd , and have multiple luns on data
rather than a big one improve performance. Does anyone have feedback about
it ?
