What are my options? How much do those cost the company?
I can run N requests on my laptop with specs of X. CPU usage hits 100% and memory usage hits 50%. This is 10% of our projected load. Therefore I project that I will need 10X CPU and 5X memory resources for this system for our entire userbase. My laptop is pretty powerful, but there are systems with 320GB of RAM and 80 cores. Do we get 2 of them?
Oh, what you have is a bunch of 3 generation old 28 core Xeons with 128gb of RAM. Can I get access to one of them to test thoroughput? One Xeon core is not as good as one macbook core, especially not that old, so I really can't make any promises of perf without testing on representative hardware. No? Fine, let's add 30% buffer for poor IPC on old hardware.
Oh, you've got a custom wrapper for Java that launches it with a bunch of custom JVM options that the SRE org mandates everyone uses? Got any documentation on what those are?
What's our lead time for getting more? We need to know that for an idea of how much buffer we need.
So apparently lead time is six months? Ok, what's our load going to be in 6 months? I'll ask bizdev how many customers they think we'll have in six months. Oh, they answered with a vague "We want to have lots of users". Fine, we'll add an extra 100% buffer.
What, the operational efficiency team are pissed our VMs are at 25% usage?
While there are certainly dev teams that throw capacity management over the wall to SRE, the inverse is also nonsense as it's the SRE team who usually get informed of hardware options, deployment standards (especially in bigco) and company operational standards.
> What are my options? How much do those cost the company? Do we get 2 of them?
What options do you want? You come to me with the specifications, and I'll get back to you if you can. I'll work with you to get what you want.
If I am the one with the power to create, the gatekeeper to the virtualization cluster and with a ballpark figure this makes my life so much easier. It allows me to make my justification easier too. "Hi Manager, X needs this. I think it needs this. I'm going to setup this and evaluate, no manager I don't think your correct."
"Here's my tech specification for what I'm setting up. here's the documentation for configuration and setup and these are the results" Lets ride.
Let me give you 8x core and 64GB see how the performance spikes and go from there. I'm not stingy and always happy to give more to test performance and than decrease if it's overkill. But don't dispute if I start to take away because of.
> what you have is a bunch of 3 generation old 28 core Xeons with 128gb of RAM. Can I get access to one of them to test thoroughput?
Sure. Yes you can. Why do you think you can't?
> Lead time
As fast as you require me to setup the VM. If I have all the docs, I can fly-by and have this thing setup under a day at minimum. Heck, I'll even work no-paid overtime to get this for you. I can escalate this on the fly. I'm in the good books with the NetOps, I even know the backup-ops. Tough crowd every time to please, but I manage so.
Production lead time? Sure, probably six months to get stake-holder approval and the rest, but I'll try to get it sooner.
Prototype lead: a week.
Just give me figures and I'll do the rest. Enterprise or not. I'll push for what you want but you have to work with me. However I need the figures and documentation before I can. I can't be seen creating the documentation on my own time based on made-up figures when I have an estate of 100 VM's needing security patches. That makes me look bad and if it fails the tests, I'll get the blame.
What are my options? How much do those cost the company?
I can run N requests on my laptop with specs of X. CPU usage hits 100% and memory usage hits 50%. This is 10% of our projected load. Therefore I project that I will need 10X CPU and 5X memory resources for this system for our entire userbase. My laptop is pretty powerful, but there are systems with 320GB of RAM and 80 cores. Do we get 2 of them?
Oh, what you have is a bunch of 3 generation old 28 core Xeons with 128gb of RAM. Can I get access to one of them to test thoroughput? One Xeon core is not as good as one macbook core, especially not that old, so I really can't make any promises of perf without testing on representative hardware. No? Fine, let's add 30% buffer for poor IPC on old hardware.
Oh, you've got a custom wrapper for Java that launches it with a bunch of custom JVM options that the SRE org mandates everyone uses? Got any documentation on what those are?
What's our lead time for getting more? We need to know that for an idea of how much buffer we need.
So apparently lead time is six months? Ok, what's our load going to be in 6 months? I'll ask bizdev how many customers they think we'll have in six months. Oh, they answered with a vague "We want to have lots of users". Fine, we'll add an extra 100% buffer.
What, the operational efficiency team are pissed our VMs are at 25% usage?
While there are certainly dev teams that throw capacity management over the wall to SRE, the inverse is also nonsense as it's the SRE team who usually get informed of hardware options, deployment standards (especially in bigco) and company operational standards.