KISS principle: stick to one thing, especially if only 1 person managing it.
On 9/30/19 9:22 PM, jh3dt68@xxxxxxxxx wrote:
Thanks Guys,
we've kicked around the idea of using Oracle VM to help manage core counts as well, but the VM admin (1 guy who manages the windows infrastructure on VMware) is not keen on dealing with another VM stack. Is it worth trying to push him into using Oracle VM ?
- Jack
On Tuesday, October 1, 2019, 12:00:40 AM EDT, Tim Gorman <tim.evdbt@xxxxxxxxx> wrote:
Hyperthreading doesn't count toward licensing.
I got 112 cores from 2 servers with 2 CPUs, each with 28 cores. 2 x 2 x 28 = 112.
I didn't add in the Oracle "core factor" of 0.5 for Intel as Jack had, so the licensable number of cores is indeed 56 vs 96.
On 9/30/19 6:59 PM, Chris Taylor wrote:
Tim,
Is this part accurate?
A new Dell R740 server running dual Xeon V2 82xx CPUs with 28 cores apiece would kick the butt off any two of those old servers, if not all four of them. Two such servers, each with 56 cores, is only 112 cores to license
2x28 = 56 but it's only 112 with hyperthreading, right? And hyperthreading isn't licenseable IIRC?
Chris
On Mon, Sep 30, 2019, 8:43 PM Tim Gorman <tim.evdbt@xxxxxxxxx <mailto:tim.evdbt@xxxxxxxxx>> wrote:
Jack,
When running any software licensed by CPU, smaller numbers of
faster CPUs are less expensive than larger numbers of slow CPUs.
What you described in your email seem to be six-year-old Dell
database servers, probably E5-2697 Xeon processors with 12 cores?
Unless I misunderstood, you mentioned four (4) servers, each with
4 sockets, each socket with 12 cores. To me, that is 4 x 4 x 12
which totals to 192, not 96, cores to license.
A new Dell R740 server running dual Xeon V2 82xx CPUs with 28
cores apiece would kick the butt off any two of those old servers,
if not all four of them. Two such servers, each with 56 cores, is
only 112 cores to license.
Updating the hardware will drop your core count from 192 to 112
while improving performance dramatically.
Oracle EE licenses with a 75% discount (i.e. list $47,500/core to
discounted $11,875/core) is a licensing drop from $2.28m to
$1.33m. If your discount is less than 75%, then savings are even
bigger. If your shop has only 15 people, I'm betting even money
that your discount isn't 75%.
Figuring that a decently loaded R740 can be bought for about
$60,000 or less, the $120k hardware purchase pays for itself with
$950k licensing savings. That'll pay for some VMware/OVM licenses
too, with enough left over for salaries for a couple developers or
DBAs.
The only way a hardware upgrade wouldn't pay for itself is if
there is a 97.5% discount (i.e. $1187.50/core is $228k for 192
cores and $113k for 112 cores, difference of only $115k). And
with a 97.5% discount with Oracle, your time is better spent on
other concerns. :)
Please re-purpose those old servers for software not licensed by
CPU, and keep the Oracle licenses on new fast hardware.
This is a golden opportunity to ditch those old boat-anchor
servers, and to ditch the boat-anchor concept of "physical
servers" as well, and build a nice VM farm for databases.
Just my US$0.02...
-Tim
On 9/30/19 4:18 PM, jh3dt68@xxxxxxxxx <mailto:jh3dt68@xxxxxxxxx>
(Redacted sender jh3dt68 for DMARC) wrote:
Hello all,
Where I work we have Oracle EE running on 4 physical Dell
servers, all running Oracle 12.2 and RHEL 7. The environments are
PROD, DR, TEST and DEV. At present all of the servers have 4
sockets with 12 cores per socket, for a total of 96 CPU cores of
Oracle EE . Unfortunately, due to budget cuts, management is
asking us to reduce our annual Oracle maintenance costs, either
that or we have to lay off a couple of developers, there are 15
people total in our shop. Our first thought was to combine DEV
and TEST as both of those environments are not fully utilized.
That would reduce our core count by 1/4. But digging into the
contracts and the world of Oracle licensing (ugh), it looks like
Oracle could re-calculate the maintenance costs based on the
current list price of annual support, not on the discount price
we received when buying Oracle 4 years ago. That means we
wouldn't pay more but maintenance costs might not be any less.
The other idea we had, was to convert the CPU licenses to Named
User Plus licenses for DEV/TEST. There are only 15 people who
ever use the DEV/TEST environments and we would leave PROD/DR
alone for now. I understand there are processor minimums which
must be accounted for, but if we combine the DEV/TEST and
switched to NUPs I'm hoping it would result in some cost
reduction, even if Oracle tries to claw back some of the savings.
Any insights or suggestions are greatly appreciated,
- Jack H.