One of the reasons why we started using Oracle VM was due to Oracle’s licensing policy with their products running on top of Oracle VM.
If you “pin” (or hard partition) the virtual machine to specific cores you don’t need to license whole server pool which the virtual machine runs on. Drawback is when you have commodity hardware with Oracle VM the only way to do CPU pinning is from the command line!
In case you have hyper threading enabled on your Oracle Virtual Server and have x86 Intel processors the licensing is quite straight forward:
VM has 8 vCPUs allocated which means you will need 2 core licenses for your Oracle product if hard partitioning/cpu pinning is used. This is due to Intel-based processor factor 0.5 and 8 vCPUs using total of 8 threads from 4 cores. 4c x 0.5 = 2. 🙂
However you will need to be careful that you have the hard partitioning in place so you don’t need to license whole server pool. Also with hard partitioning in use you will not be able to use live migrate for your VM.
The actual CPU pinning with 3.4.2
I want to check what my OVS has in terms of vCPU’s I can allocate to one VM:
[root@OVS1 ~]# xm info host : OVS1.local release : 4.1.12-61.1.9.el6uek.x86_64 version : #2 SMP Tue Sep 13 22:53:36 PDT 2016 machine : x86_64 nr_cpus : 32 nr_nodes : 2 cores_per_socket : 8 threads_per_core : 2 cpu_mhz : 2700
In this server I have 2 CPU sockets with 8 cores each and 2 threads per core. This adds to total of 32 CPUs.
Next I want to check which VM’s are using which CPUs. As seen part of CPUs are allocated to dom0 already:
[root@OVS1 ~]# xm vcpu-list Name ID VCPU CPU State Time(s) CPU Affinity Domain-0 0 0 0 -b- 36760.2 0 Domain-0 0 1 1 -b- 26625.9 1 Domain-0 0 2 2 -b- 46877.3 2 Domain-0 0 3 3 -b- 20995.3 3 Domain-0 0 4 4 -b- 38292.6 4 Domain-0 0 5 5 -b- 21067.0 5 Domain-0 0 6 6 -b- 89754.7 6 Domain-0 0 7 7 -b- 75418.8 7 Domain-0 0 8 8 -b- 47228.8 8 Domain-0 0 9 9 -b- 29581.8 9 Domain-0 0 10 10 -b- 35139.4 10 Domain-0 0 11 11 -b- 20257.9 11 Domain-0 0 12 12 -b- 27636.9 12 Domain-0 0 13 13 -b- 17393.8 13 Domain-0 0 14 14 -b- 27594.6 14 Domain-0 0 15 15 -b- 16195.1 15 Domain-0 0 16 16 -b- 97024.1 16 Domain-0 0 17 17 -b- 72472.6 17 Domain-0 0 18 18 -b- 86671.8 18 Domain-0 0 19 19 r-- 125745.7 19 0004fb0000060000c04cfa6f302883fa 23 0 21 -b- 4074.3 16-31 0004fb0000060000c04cfa6f302883fa 23 1 31 -b- 2706.9 16-31 0004fb0000060000c04cfa6f302883fa 23 2 23 -b- 1013.0 16-31 0004fb0000060000c04cfa6f302883fa 23 3 22 -b- 4148.8 16-31
I’m interested in VM 0004fb0000060000c04cfa6f302883fa named test_vm_1 which uses CPUs (threads) 21,22,23 & 31 currently. I have downloaded OVM tools which are found in patch 13602094 so I can use ovm vmtools to check current status:
[root@OVM1]# ./ovm_vmcontrol -u admin -p password -h OVM1 -v test_vm_1 -c getvcpu Oracle VM VM Control utility 2.1. Connecting to OVM Manager using Web Service. Connected. OVM Manager version: 3.4.2.1384 Command : getvcpu Getting pinned CPU list... No CPU pinned for VM 'test_vm_1'
At the moment CPUs are not pinned. In Oracle VM release 3 you can use OVM tools to do the actual pinning:
[root@OVM1]# ./ovm_vmcontrol -u admin -p password -h fOVM1 -v test_vm_1 -c setvcpu -s 21-24 Oracle VM VM Control utility 2.1. Connecting to OVM Manager using Web Service. Connected. OVM Manager version: 3.4.2.1384 Command : setvcpu Pinning vCPU '21-24' to VM 'test_vm_1' Pin vCPU succeed. [root@OVM1]# ./ovm_vmcontrol -u admin -p password -h OVM1 -v test_vm_1 -c getvcpu Oracle VM VM Control utility 2.1. Connecting to OVM Manager using Web Service. Connected. OVM Manager version: 3.4.2.1384 Command : getvcpu Getting pinned CPU list... Current pinned CPU:21-24
But now after rebooting the VM with OVM 3.4.2 the CPUs are not getting pinned. This can be seen running the vcpu-list command:
[root@OVS1]# xm vcpu-list Name ID VCPU CPU State Time(s) CPU Affinity 0004fb0000060000c04cfa6f302883fa 23 0 23 -b- 8.7 16-31 0004fb0000060000c04cfa6f302883fa 23 1 28 -b- 2.0 16-31 0004fb0000060000c04cfa6f302883fa 23 2 31 -b- 1.9 16-31 0004fb0000060000c04cfa6f302883fa 23 3 21 -b- 1.6 16-31
In OVM 3.4.X I still need to login to actual physical OVS and run xm vcpu-pin which wasn’t required earlier with 3.3.X. After that the CPUs are pinned and CPU affinity shows up correctly:
[root@OVS1]# xm vcpu-pin 23 all 21-24 [root@OVS1]# xm vcpu-list Name ID VCPU CPU State Time(s) CPU Affinity 0004fb0000060000c04cfa6f302883fa 23 0 24 -b- 9.4 21-24 0004fb0000060000c04cfa6f302883fa 23 1 21 -b- 2.1 21-24 0004fb0000060000c04cfa6f302883fa 23 2 23 -b- 2.3 21-24 0004fb0000060000c04cfa6f302883fa 23 3 22 -b- 1.7 21-24
That’s it! You can also review your VM’s vm.cfg and you should see the CPU pinning in place there as well.
If you would use Oracle’s Private Cloud Appliance which is sort of pre-build Oracle VM you would have “trusted partitions” in use. Then you wouldn’t need to do this manually but all what would be required is OEM connection which Oracle can use (scary huh?) and you would be set. If there would be possibility I would definitely be looking to migrate from “standard” Oracle VM to PCA. However PCA’s Oracle VM version is bit behind of current 3.4.2..
Couple Oracle’s documents also on hard partitioning which are useful and have above in more detail:
Set CPU Pinning for VMs on Oracle VM 3.4.x (Doc ID 2213691.1)
Insightful! Great article..!
Will HA feature be restricted after CPU pinning? For example, CPU 0 -7 are pinned by test-vm1 in OVM server 1 which encounter hardware failure. Could test-vm1 be able to failover to server 2 which CPU0-7 are being occupied?
Yep that will be restricted, so you can’t use live migration if you use CPU pinning. https://docs.oracle.com/cd/E50245_01/E50249/html/vmcon-vm-pinning.html
Hi Simo,
Right, the live migration is restricted if I use CPU pinning. However, I would like to know if HA cluster feature among OVM servers is also restricted .
https://docs.oracle.com/cd/E50245_01/E50249/html/vmcon-svrpool-ha.html
It seems to be two different OVM features. Correct me if it is wrong concept.
Looking forward to seeing your advice.
Yeah I was thinking it in terms of uncontrolled failure and when VM will be automatically migrated (even though it would be shutdown) to a different VM server. You wouldn’t have the CPU pinning in place if the VM would be automatically started and would need to pin it manually. So it would be a very thin line what you would need to license/wouldn’t need to.
So since you would need to do it manually afterwards it wouldn’t really be a real HA configuration anymore? 🙂
Question:
I have 2 CPU database licence and VM server:
nr_nodes : 4
cores_per_socket : 4
threads_per_core : 2
——————————–
CPU core socket node
CPU0 0 0 0
CPU1 0 0 0
CPU2 3 0 0
CPU3 3 0 0
CPU4 8 0 0
CPU5 8 0 0
CPU6 11 0 0
CPU7 11 0 0
CPU8 0 1 1
CPU9 0 1 1
CPU10 3 1 1
CPU11 3 1 1
CPU12 8 1 1
CPU13 8 1 1
CPU14 11 1 1
CPU15 11 1 1
CPU16 0 2 2
CPU17 0 2 2
CPU18 3 2 2
CPU19 3 2 2
CPU20 8 2 2
CPU21 8 2 2
CPU22 11 2 2
CPU23 11 2 2
CPU24 0 3 3
CPU25 0 3 3
CPU26 3 3 3
CPU27 3 3 3
CPU28 8 3 3
CPU29 8 3 3
CPU30 11 3 3
CPU31 11 3 3
—————————-
Name ID VCPU CPU State Time(s) CPU Affinity
0005fb0000050000c00a3da67855a32b 15 0 1 r– 9.8 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 1 9 r– 9.7 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 2 2 r– 9.4 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 3 9 -b- 9.8 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 4 – –p 50.3 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 5 – –p 50.2 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 6 – –p 50.2 1-2,8-9
0005fb0000050000c00a3da67855a32b 15 7 – –p 50.3 1-2,8-9
0005b4353454445678ah5d473643db21 10 0 5 r– 200.9 3-5
0005b4353454445678ah5d473643db21 10 2 3 r– 200.7 3-5
0005b4353454445678ah5d473643db21 10 3 – –p 200.1 3-5
0005b4353454445678ah5d473643db21 10 1 4 -b- 200.7 3-5
0005b4353454445678ah5d473643db21 10 4 – –p 200.2 3-5
0005b4353454445678ah5d473643db21 10 5 – –p 200.2 3-5
Domain-0 0 0 0 -b- 200.4 0
Domain-0 0 1 1 -b- 200.0 1
Domain-0 0 2 2 -b- 200.6 2
Domain-0 0 3 3 -b- 200.2 3
Domain-0 0 4 4 -b- 200.0 4
Domain-0 0 5 5 — 200.5 5
Domain-0 0 6 6 -b- 200.7 6
Domain-0 0 7 7 -b- 200.8 7
Domain-0 0 8 8 -b- 200.3 8
Domain-0 0 9 9 -b- 200.2 9
Domain-0 0 10 10 r– 200.5 10
Domain-0 0 11 11 -b- 200.4 11
Domain-0 0 12 12 -b- 200.5 12
Domain-0 0 13 13 -b- 200.6 13
Domain-0 0 14 14 -b- 200.6 14
Domain-0 0 15 15 -b- 200.3 15
Domain-0 0 16 16 r– 200.8 16
Domain-0 0 17 17 -b- 200.4 17
Domain-0 0 18 18 -b- 200.4 18
Domain-0 0 19 19 r– 200.6 19
0004fb0000060000d684503452353224 7 0 12 -b- 300.8 12
———————————————
Is it correct way to pin two cpus like a this:
./ovm_vmcontrol -u admin -p somepassword -h $(hostname) -v vm_machine_name -c setvcpu -s 20-23,28-31
and what is corrent number of cpus to give to this “vm_machine_name” through VM Manager GUI, max-min=2 cpus, or max-min=8 cpus as it should represent vCPUs (vCPUs~threads)?
Hi, Nice Blogs.
Here ‘s my question. If I Do Not Pin (Hard Partition) cpus to a vm_guest and my bare metal server has 8 cores. Does it mean I have to license Oracle EE for 8 cores? Btw this is the only vmguest on the server pool
Yeah correct, then you license all the cores regardless how many vcpus you have assigned for the VM. Then again if you license them all you can run as many EE VM’s in that one physical server as you want.
Hi,
How tu upgrade OVS servers with pinned guest on it with no downtime. So when i Live migrate mine pinned guest to another host it is pinned there to the same CPU-s.
Can I LM my guest and after OVS upgrade is done move them back? If not how to do it?
Thanks
Z
Hi,
I am trying to figure out how to determine the minimal number of licenses of the Oracle Internet Application Server Enterprise Edition (IASEE) linked to my EBS server.
From what I understood so far, to determine the minimum number of IASEE licences I have to use the following formula:
npl * coeffcpu * nblminias
where:
npl is the number of “Processor” licenses calculated from the EBS server characteristics.
coeffcpu is the multiplying coefficient that depends on the CPU type (e.g.: 0.5 for Intel CPU)
nblminias is the minimum number of licenses imposed the Oracle product IAS.
What I struggle with is how to determine npl for my EBS server!
The EBS server is a VM in a Oracle VM environment and this VM relies on an hypervisor that has a total of 24 vCPU.
The Hard Partitioning and CPU pinning are activated.
Is the following true?
If I allocate/pin 4 out of the 24 vCPUs to my EBS VM then npl is 4?
Cheers.
Hi,
Thank You for Your great article.
I’am curious: is it possible to bind certain vCPU in users’ virtual machine with certain CPU core/thread ( 1 to 1, like OVM doing it only towards “system VM ” Domain-0 ) ?
How to do it ,if it is possible ?
Regards