That blog post is good, but it is over built in my opinion, all you really need is a single daily sum of all spend across your account and chart that. Maybe a stacked column with product color or something. This will allow you to quickly see if one product consumption is shooting up unexpectedly at a glance vs checking something at the end of the month and working with your account reps to figure something out.
6k z z
5k z z
spend 4k z z z
3k z z z
2k x x x x x x x x x
1k x x x x x x x x x
date 1 2 3 4 5 6 7 8 d+
x = expected
z = unexpected
For example, maybe you left a test env running or something and all of the sudden you're spending an extra few K per day. Maybe you're doing some query loops over a large BQ dataset and racking up tons of $$. Put the chart up on your big TV screen in the Ops area so everyone can quickly check it with all your other metrics. When you're spending 100k+/month, with a few folks digging around the console, it's hard to notice an extra 5k-10k here or there. This quick tip will totally help you visually sanity check if something is off. So, this might not be a killer solution for folks spending a few thousand a month but will totally saves tens of thousands as you scale up.[1] https://cloud.google.com/billing/docs/how-to/export-data-big...
[2] https://medium.com/google-cloud/visualize-gcp-billing-using-...
1: https://cloud.google.com/resource-manager/docs/creating-mana...
And one of the advantages of having a zillion different products approach is that you can trust AWS to grow with you and improve their products over time. Same can't be said of GCP.
Compute is compute. Both providers offer similar instance classes for the most part. GCP offers sustained-use discounts, in addition to committed use discounts that match up to reserve instances, plus billing by the minute.
Spot instances can be hugely variable, but cheaper than prremptible. Preemptible offers predictable costs with a guarantee that the instance will go away every 24 hours (and come back).
Network IO is way better.
Inter-region networking comes for free.
GCP ssd persistent disk lands between AWS GP2 and IO1 and gives way more predictable performance.
Local disks can be attached to any instance and end up giving similar performance and being way cheaper than I2 and I3 instances.
Wait until the next post comes up about “serverless” and someone mentions that there are servers involved....
There is no guarantee that the instance will come back. And it can be shut down in under 24 hours. Not that that's a problem compared to AWS or Azure.
>Because preemptible VMs have no availability guarantees, you should design your system under the assumption that any or all of your Compute Engine instances might be preempted and become unavailable. There are no guarantees as to when new instances become available.
https://cloud.google.com/kubernetes-engine/docs/how-to/preem...
There's also reOptimize for in-depth reporting with a solid free plan: https://www.reoptimize.io/
Azure's VMs are definitely overpriced compared to both GCP and AWS, but most other services are reasonable.
Not sure what you mean about Azure's "strange UI"? Personally, I find the Azure Portal to be far more consistent than AWS' or GCP'. I also think it looks and feels great, and I'm obviously not the only one - I literally have customers asking me to replicate the Azure UI for their web apps!
Lastly, I find it somewhat incredible that you would complain about Azure's tooling - I've found it to be excellent: a snappy web UI, cross-platform Powershell, cross-platform Azure CLI, REST API, ARM templates... seriously, I'd be interested to hear what your beef is with the tooling?
You need to drill down to the network card primitive, figure out which subsection you need to configure and applying the same rule is again an exercise in clicking.
In GCP: Apply a tag. Create firewall rule that targets the tag.
But then again, this is higly subjective and I do understand that somebody would prefer Azure way.
Re: tooling, when I was using CLI was in both Node and Python, docs were lacking and the strange Resource Manager vs. Legacy had so many quirks we had to run windows to get proper powershell for configs of VMs.
Leader in k8s? Not for production. I defy you to do a gcp gke deployment without using beta or alpha features (that specifically say don’t use this for production). Stackdriver error reporting springs to mind. And working with SSL certs/let’s encrypt is a close second.
If you want a couple more peeves for your list how about documentation? I regularly find documentation heavy on theory light on specifics.
GCP is my favorite thing that is almost good enough to use. I’m hoping the change of leadership actually improves things. GCP could stand to re-focus on a “customer first” philosophy.
I don't think we use any Alpha features, I know there are a couple Beta APIs from Kube we use, but we would be using those no matter where we ran our Kube deployment. We use fluentd for our log collection and Prometheus for metrics and alerting combined with Elastalert. So don't have any experience with Stackdriver.
Why is working with SSL certs an issue? We use the NGINX Ingress with cert-manager and have yet to have an issue. We only have a couple domains pointed at the cluster though, not sure if thats where your issue comes from. The two just worked though with very little tweaking.
Average availability for the year across those environments is 99.99%.
I'd say that's decent enough to use.
Their support is hilariously bad at times, though.
I find it a bit ironic that the mobile and web side of the house are pushing very heavily for IPv6 adoption and yet they expect their cloud customers not to need it.
That was my experience when we needed something.
Dont get me started on docs and operations.
Eg: Using Azure we found out the hard way that the Firewall was cutting connections after 5m of inactivity. Changing that setting? Their docs are wrong, stack overflow kindof helps, but to fix it we had to use windows based tooling in a vm since their linux based stuff didnt work.
I know this will happen sooner or later with anybody, but on average the experience with GCP was superb.
I do agree re: certs. That is the only thing I really hate, doing LE when they could run this for you. But they will launch an SSL solutio since they have one now for AppEngine.
Could you elaborate on this? Which firewall is this? I remember something similar with their load balancer but I thought this was fixed years ago. Might you have some link to the issue? Thanks.
StackDriver error reporting is a lite product intended mostly for folks with small applications on AppEngine.
There is a recent post by Corey Quinn entitled CloudWatch Is of the Devil, but I Must Use It discussing the extremely poor usability of Amazon CloudWatch [0]. Here’s another story about the general weirdness of the product [1]
I would strongly recommend using a dedicated exception tracker like Sentry.io, they have an excellent full featured product at a extremely reasonable price.
GCP has clearly documented Platform Launch stages for Alpha, Beta, and General Availability [2].
The Kubernetes project itself follows the same definition for these identifiers as GCP.
Specifically Beta is considered to be acceptable for most production use-cases, but it is not guaranteed that the API will remain backwards compatible.
So if you’re running the majority of cloud services, beta is fine. If you’re running a system that will cause permanent harm to humans, don’t use beta.
One last note, GCP has just released a fully managed certificate product for their global load balancer. There is fully integrated Kubernetes support which has not yet been announced formally [3].
Certificates for Kubernetes on Amazon are less integrated than the above. There’s several options, it’s not clear without extensive research and testing what the tradeoffs are, and in the end it turns out that there is not an officially supported Amazon option.
[0] https://www.linuxjournal.com/content/cloudwatch-devil-i-must... [1] https://www.circonus.com/2016/10/no-fixed-glitch/ [2]https://cloud.google.com/terms/launch-stages [3] https://github.com/GoogleCloudPlatform/gke-managed-certs
I used to do $100,000 a month in revenue on Amazon with one HP skew. The product grew to #1 in its category within a couple months, another month after which Amazon started selling it to, "shipped from sold by Amazon".
Amazon is happy to help be your growth partner until you prove worth devouring.
AWS is their most profitable business in terms of margin. It makes sense to avoid funding your competition if possible, which is a problem for Amazon given how many business divisions they have and continue to add.
Based on my own experience as the CTO: After being an Azure shop for a year we've migrated ~50-100 VMs to GCP and I love the GCP products.
GCP is:
- Simpler to use
- More tailored to people with Linux environment
- Leader in K8S
- Has good support
- So much cheaper (in our case we saved ~60%)
- Has great UI and understandable primitives.
My only pet peeve is the fact that exporing your spend is practically impossible unless you're a BQ guy that can work directly with report exports.
PS: We're building our future infra on K8S to allow us to migrate more easily to a different could if something goes awry with GCP, I really hope there won't be a need to migrate back to Azure and its arcane and high pricing, strange UI, worse tooling...