[thelastgallon]

> It's largely because a lot of Oracle DB products where performance mattered (eg. Exadata) needed some sort of a base OS that Oracle could manage and optimize as needed

All that’s needed is update sysctl.conf to tune kernel parameters to the workload. Every Linux sysadmin knows how to do this. What kernel parameters need to be updated is heavily documented for any product.

[alephnerd]

Having these pretuned AND having a dedicated support team doing the tuning for you is the killer app for some buyers, as it allowed Admins to concentrate on other work and also minimize management overhead for the Infra org.

Spending $500k/yr on compute+support SLA is cheaper than $200k/yr on compute and hiring 3 admins dedicated to that piece of compute.

This is the model that every Enterprise Infra vendor pushes (eg. Oracle, AWS, MongoDB, Nvidia), and most mid- and upper-market purchasers are used to it.

[thelastgallon]

> Having these pretuned AND having a dedicated support team doing the tuning Dedicated support team costs a lot more than you think. I guess you never had to deal with Oracle support. If you get hold of someone, all they will do it point to the documentation.

All software products have documentation on how to install the product. Oracle has a large suite of products, their databases, ERPs, etc. For kernel parameters, its just a file, which takes a second https://docs.oracle.com/en/database/oracle/oracle-database/1...

In reality though, all Infra teams, have infrastructure to install OS (and manage the fleet), then post-install customize the OS to which team is requesting, usually done via puppet or ansible to manage the configuration. There will be standardized configuration for application, web, database (just to keep it simple).

I would be shocked if Oracle support (or any other vendor) is given login access to make changes on servers owned by clients. At best, you open a case, you get an incompetent support person who'll send you documentation.

Oracle support does not replace admins. Oracle support gives you access to bug fixes, updates, documentation. I believe you can download most Oracle software for free, but without the docs and updates, its worthless. Other vendors may use the opposite strategy, docs openly available but software downloads are paid/subscriptions.

> Spending $500k/yr on compute+support SLA is cheaper than $200k/yr on compute and hiring 3 admins dedicated to that piece of compute.

In reality though, there will always be admins, then a whole lot DevOps/Cloud Ops/Kubernetes/SRE/etc people added, smooth talking manager/director increasing the spend from what could be done on bare-metal under 20K to a 20 million dollar multi cloud strategy. Why have 3 admins report to you, when you can have an army of 200 people do the same work for 100x more cost? Success stories and promotions all around!

[alephnerd]

> all Infra teams, have infrastructure to install OS (and manage the fleet), then post-install customize the OS to which team is requesting, usually done via puppet or ansible to manage the configuration.

Yep! And it takes time and effort to maintain your Puppet/Chef/Ansible/Terraform/OpenTofu scripts as well as your golden images as well as triaging escalations as well as other incidental work. This means you don't have as much time to work on tuning or debugging, because you'll have dozens of tools (some in-house, others purchased) to manage.

Furthermore, most people recognize Hardware specialized IT Administration is increasingly a career dead end, so most end up switching to Engineering, Sales Engineering, or Support Engineering due to better career opportunities.

> I would be shocked if Oracle support (or any other vendor) is given login access to make changes on servers owned by clients. At best, you open a case, you get an incompetent support person who'll send you documentation.

This is the norm in most mid- and upper-market support contracts. You'll have a dedicated TAM, Support Eng, and CSM who will handhold teams, and will have access to the underlying infrastructure.

> Oracle support does not replace admins. Oracle support gives you access to bug fixes, updates, documentation. I believe you can download most Oracle software for free, but without the docs and updates, its worthless. Other vendors may use the opposite strategy, docs openly available but software downloads are paid/subscriptions.

Depending on your contract, you would be given a dedicated TAM team and support team to debug any issues in the Oracle stack.

> In reality though, there will always be admins, then a whole lot DevOps/Cloud Ops/Kubernetes/SRE/etc people added, smooth talking manager/director increasing the spend from what could be done on bare-metal under 20K to a 20 million dollar multi cloud strategy. Why have 3 admins report to you, when you can have an army of 200 people do the same work for 100x more cost? Success stories and promotions all around!

That "smooth-talking manager" needs to justify to the CFO, COO, CTO, VP Eng, etc that for $X spent, I can get 1.5 * $X back.

As I've mentioned on multiple different occasions on HN, spend on on-prem infra is treated as part of the Finance+ITOps budget, not the DevOps budget (which is generally within R&D).

Procurement is hard, and you need to JUSTIFY a 1% increase in headcount

For example, let's assume you are hiring 3 IT Admins for $120k. That ends up costing $700-800k/yr because of benefits and incidentals. The compute as well is an additional $200-300k.

This means you are spending $900k/yr AT BEST.

That $200-300k in compute becomes $500k with a support contract, and you can hire 1 person for $120k to manage that.

This means you're spending around $750k/yr AT BEST.

That extra $150K can then be given to Engineering to help give bonuses to attract good dev talent or hire some additional headcount on the Sales side to sell the product you are hired to build.

[twoodfin]

MongoDB offers a preconfigured Linux to run on?

Or did you just mean Atlas?

[alephnerd]

I mean MongoDB has a professional services SKU to simplify onboarding and management.

[johannes1234321]

> All that’s needed is update sysctl.conf to tune kernel parameters to the workload.

Mind that they do quite some work on the kernel itself to optimize it for their workloads:

https://blogs.oracle.com/linux/post/oracle-is-the-1-contribu...

The availability of Oracle's uek kernel is a differentiator from standard RHEL.

[zaphar]

Oracle has a long history of not documenting all of this stuff and instead suggesting you should hire one of their army of consultants to help tune the OS or Database for you.

[mingus88]

If we are talking about tuning for a specific workload, what’s wrong with that?

If your in-house DBA doesn’t have the experience to perform the specific tuning required, then that’s what support contracts are for

The documentation can’t cover every customer’s use case and configuration. That’s just enabling folk to blindly copy inappropriate sysctls they don’t understand like they are building gentoo kernels.

[zaphar]

You don't need to cover every use case. You just need to document what stuff does and how it affects to various workloads. A good engineer can from that information infer what they need for their workload. But not documenting that a control exists so that you can be bill you for consulting puts you in my "will not use" category.