Hello ladies and gents of, I have a simple question: What is the minimum amount of Oracle licensing that one needs to purchase for AWS Dedicated Host that runs M4.Large instances?. This dedicated host will have two physical sockets. I know it's going to be two Xeon chips, but not sure which model.
![Licensing Licensing](/uploads/1/2/5/4/125430685/526470043.jpg)
![Aws Aws](/uploads/1/2/5/4/125430685/855530855.jpg)
When you run Oracle Database on Amazon AWS you Bring Your Own Licenses depending on the number of virtual cores (which are the number of cores allocated to your vCPUs). Behind the instance types, you have different processors and hyper-threading.
It can host up to 22 m4.large instances (each instance is 2vCPUs/8 GB of RAM). So we're talking 44 vCPUs, which should be 22 physical cores, which means that we're talking 11 cores per cpu?
This seems odd to me, I would've expected 12-core per CPU. We will end up running all 22 VMs at some point in the next 2 years. So assume full capacity (not that it matters for licensing). I am betting that we need two Oracle SDE2 licenses, but I need to hear from someone who's been in the same situation. My reasoning: each SDE2 is for one physical CPU with artificial limit of 16 cores per CPU.
Since we're 2 CPUs and 22 cores, we should be fine. Unless Oracle thinks we're 2 CPUs and 44 cores? In which case, bummer. If you have no info, do you know whom to ping on AWS side who might be able to help? It's an uncommon question and we found that even their technical staff is sometimes not clear on dedicated hosts. We tried through Oracle support, you can guess how that went. I will cross-post this to too.
Many thanks in advance,. T-Hex. Each thread is definitely not a socket. Socket is a physical chip and with SDE 2 you get up to 16 real cores per socket. Source for this claim: There are several problems with the above document: (1) It talks about EC2 instances running in AWS cloud.
This is different than EC2 instance running on your dedicated host. You are not a part of AWS when you are on dedicated host (for legal purposes). (2) They consider each chip (i.e. A socket) as running a max of 16 physical cores. This was fine up until recently, because it was known that each vCPU in EC2 was only running half a core. So on 16 physical cores, you could spin up 32 vCPUs.
If the story stopped here, we'd be great: Dedicated host that we want allows up to 22 EC2 instances, each with 2 vCPUS. In real world that's 44 vCPUs, which we know are only 22 physical cores. And this would fit in two SDE 2 licences. But what worries me is that they now appear to count this as 44 physical cores (i.e.
1 vCPU = 1 core) in EC2 world. Thus my original post: how does this work in dedicated host model, which should not be the same as EC2-in-AWS-cloud model. Infuriating, I know:(. IANAL but if your interpretation is correct, it sounds like you found a loophole is Oracle's wording of the licensing. I'm 99% sure Oracle's intention is to charge you per thread, because the whole point is to make running on AWS hardware prohibitively expensive for people that can't manage their own servers but won't use Oracle's cloud offering. If their license lets you still do it affordably on a dedicated ec2 instance, and you push them to acknowledge it, their response will likely be to close the loophole. But of course if you don't push them to acknowledge it and start using it somehow, they might take action against you.
Read your Oracle license carefully. Everyone got scared when Oracle put this out: But read the fine print on the bottom. It's a guideline only and does not form part of a contract. Notice how the AWS RDS pricing model for Oracle hasn't changed since Oracle released this. If you had a licensing contract with Oracle prior to this being released it is very likely the terms on that contract have not changed, so the old model of mapping vCPUs to core licenses may still apply. Again, read your license carefully and don't trust the Internet for advice on your specific license terms (including this post).
There are whole companies established to help organisations navigate the practical application of Oracle licensing, using them could save your organisation millions:. Just want to chime in: we had 100% feature compatibility pass when we migrated one of our Oracle databases onto Aurora PostGres.
The problems we faced from that point on were political:. our clients use Oracle only. They have no staff for Postgres. our own staff would have to retrain. The second point (retraining) is something that would be fine by me. But it would be impossible to sell the clients on a DB back-end that they don't know. If your business model is such that you can host and will never have to sell the app to be client-hosted, go for Aurora at warp speed 9:).
The last I heard from Oracle sales person was that for dedicated host, you must licence 2 SDE2 licences (one licence per socket, since dedicated hosts have a min of 2 sockets). If you have a dedicated host with more than 2 sockets, you can't use SDE2 and probably must use enterprise (I didn't pursue this, since we are looking for 2 socket servers only). Here are some kinks though: Oracle artificially caps you at 16 hyper-threads total on those two sockets (combined). It doesn't matter that AWS offers more than 8 cores per CPU, you can't use them. Oracle still counts vCPUs in amazon as half a thread.
So your m4.large (2 vcPU, 8 GB of RAM) counts as using one hyper-thread. So you can have up to 16 EC2's with Oracle on dedicated host. Now this last part still needs verification. The Oracle person was mumbling here and first stated that you can only have 4vcpus (for some reason) and seemed to imply that you can only have 2 EC2s?
We started pulling words out of his mouth and in the end, in principle, he agreed that 16 EC2s is a possibility. He will continue to check on this. Meantime, we're still waiting on someone from AWS to get in touch with us to discuss all of this. If you want to get your guys involved, the one question that needs to be answered is this: how much licensing do I need on an m4.large class of a dedicated host to run all 22 VMs on it?
It would appear like your upper limit is 16 VMs due to licensing. And that's if we're lucky with Oracle interpretation.
Completely fucking absurd, in 2017.