GCP offering 200Gb free egress on their Standard Tier | Hacker News Reader

GCP offering 200Gb free egress on their Standard Tier

48 pointsPaywallBuster27 comments

[received by email]

We are writing to let you know that starting October 1, 2023, all existing and new Standard Tier Egress customers will avail their first 200 GB for free in every region that they choose to operate, after which Standard Tier will be billed at current rates.

What do you need to know? The 200 GB meter will reset every month. We’re introducing this offer to make it frictionless for customers to start using Standard Tier egress with their Virtual Machines.

Today, there is no free usage tier for Standard Tier Egress. With this MSA, we are introducing a free tier of egress to our Standard tier customers.

What do you need to do? No action is required on your part. However, if you currently consume Standard Tier Egress, you will notice a slight reduction in the billed amount starting October 1, 2023. You can also review our standard billing rates.

Thanks for choosing Standard Tier Egress

Still won't use gcp. VPS like digital ocean and linode are the way to go

If cloud compute is an “Xbox vs. PlayStation” or “iPhone vs. Samsung” type of situation, you are obviously not the type of customer that GCP even tries to attract. Cloud is for big computational needs that can't be covered by Linde or DO.

yeah and gcp is definitely not for that either. they had a full region down for weeks in paris while pretending they have availability zones (but one fire in one datacenter put everything down). at this stage the only trustable big computational cloud is AWS.

wasn’t that OVH?

I think they're referring to this incident?[1] although the timeline in the preliminary report seems to indicate that service disruption in europe-west9 but outside the directly affected zone had ended the day after the incident began. In the directly affected zone, there was water and smoke damage, and it seems like 58% of europe-west9-a was decomissioned as a result. Losing roughly half of the capacity in one of three availability zones for a location would probably make instance availability a lot tighter, and might have impacted the poster in ways that aren't reflected in the report?

I work for a gcp customer, but we don't have anything deployed to europe-west9, so I don't have direct exposure. As I recall, we did see some difference in geodns routing on that day, but it was unclear if it was related or not.

[1] https://status.cloud.google.com/incidents/dS9ps52MUnxQfyDGPf...

definitly not one day but weeks as I posted links just above. and again they lied about what is a region. don't know why it didnt reach WW news.

The register link says europe-west9-a was still down after weeks, and that -c had a five hour outage the weekend before publishing. The other article was published near the time of the incident and doesn't say anything about the weeks after.

europe-west9-a was clearly messed up, and for a long time; which I think is fairly reasonable if you have to toss out at least 58% of the hardware because of water and smoke damage. c'est la vie. That -b and -c were unavailable during the incident is not how availability zones are supposed to work, and it's bad; but you seem to be claiming that -b and -c were down for weeks afterward and I don't see that reflected in the links you shared.

If you demand a 3 availability zone system, and one availability zone had a disaster; the right thing to do is to move to another region which has three working azs (and hope that one is properly setup; or request an audit if you've got leverage). Cloudyness is supposed to make it easier to move infrastructure. But if likeminded customers are all moving infrastructure, there's going to be a capacity crunch. OTOH, GCP has several regions in europe-west, and there's a lot of interconnections in Europe in general, so unless you had a requirement to domicile data in France, lots of options.

https://status.cloud.google.com/incidents/dS9ps52MUnxQfyDGPf...

Multiple Google Cloud services in the europe-west9 region are impacted.

'Water intrusion in a data center in europe-west9 caused a multi-cluster failure that led to a shutdown of multiple zones. We expect some unavailability in the europe-west9 region. There is no current ETA for full recovery of operations in the europe-west9 region at this time. We expected to see an extended outage for some services. Customers are advised to failover to other regions if they are impacted.'

can you please explain to me how water intrusion in a data center cause a region failure? there is only one reason. they lied. most likely overhyping the region resiliency. how after seing that can you trust them to host critical apps is beyond belief.

and if you look the resolution link I shared they kept hiding reality behind fuzzy language.

'After Regional Spanner, IAM and GCE/PD came back online for the region, almost all of the services in the region fully recovered soon after that. The following services either took longer to recover or had additional impact during the outage.

Google Cloud Storage, Google BigQuery, Cloud Bigtable, Cloud SQL, Cloud Pub/Sub, Google Cloud Dataflow Please log a case with our Support team if you would like to receive additional information regarding these services.'

please look the last sentence and I can tell you it was way more than a day for those services. try to get additional info on how long bigquery/bigtable were offline.

> can you please explain to me how water intrusion in a data center cause a region failure? there is only one reason. they lied. most likely overhyping the region resiliency. how after seing that can you trust them to host critical apps is beyond belief.

Clearly, they fucked up, and didn't distribute critical infrastructure adequately across the region. There's lots of ways this kind of thing happens. I've had 'redundant fiber' that was running in the same bundle and got severed by the same backhoe, for example. The easiest way to determine if there's a dependency on a particular building is to forcibly de-energize all the equipment in it. In this particular case, it seems like there was a mistake in allocation of instances for the regional Spanner database, resulting in that database becoming unavailable when the building was de-energized; and then lots of services depend on regional Spanner, because that's its intended use.

It sounds like they did an audit of regional Spanners after this incident, but who knows what other hidden dependencies there are there, or at any cloud provider. Having your critical apps available in multiple distributed datacenters is best practice; having that all in one city isn't really. In an ideal world, you're not dependent on any single entity you don't control; but egress charges make that difficult.

Re: Bigquery, there's a separate post for that [1] which seems to indicate "The issue with Google BigQuery has been resolved for all affected users as of Wednesday, 2023-04-26 17:05 US/Pacific." The incident began around "26 Apr 2023 06:25 PDT", so resolution was next-day (but that's about 35 hours). If you've got other data on BigQuery, that's fine; I'm not saying your experience is wrong, I'm saying it doesn't match the public record.

[1] https://status.cloud.google.com/incidents/TbcwMSkKy8MTmeeEiq...

its not only spanner multiple services got hit. the console wasnt even accessible. to me it looks like they have one datacenter and in this datacenter multiple "availability zone"

Yeah, because the console and IAM auth rely, and pretty much everything at GCP relies on regional spanner; that's a key point of GCP design for better or worse. It's like if they only had BGP in the one datacenter. If that datacenter turns off, nothing works in the other two that are nearby.

https://www.theregister.com/2023/05/10/google_cloud_paris_ou...

https://thenewstack.io/paris-is-drowning-gcps-region-failure...

Can you please link to the incident where the "full region" was down for weeks?

I like AWS more and use it for my projects but no one can pretend Azure doesn’t exist. I would rather give them my business than GCP.

Also,one cannot pretend that AWS does not force customers to pay through the nose for its cloud services.

And, AWS offers a 5$/month VPS with 1 or 2 TB free egress traffic, coupled with affordable S3 with free egress, in a product called Lightsail.

Honestly, it’s better than DO and Linode.

> at this stage the only trustable big computational cloud is AWS.

Didn't AWS had in the last year two or three large scale events that paralyzed the web?

Edit: here's a discussion over one of the times AWS's us-east-1 was down this year.

https://news.ycombinator.com/item?id=36315300

> Cloud is for big computational needs that can't be covered by Linde or DO.

I don't think this is true at all. The cloud is just a fancy word for using someone else's computational resources, and the cloud's main selling points is not having to host or manage your hardware, and be able to scale on demand. A third less mentioned topic is being able to have global deployments.

None of these items imply a limit on how "big" are your computational needs. In fact, it's well established that once a company grows its cloud needs after a threshold, it becomes far more cost-effective and performance to just manage your own hardware and infrastructure.

Lastly, even AWS frames itself as the go-to solution to grow from zero to launch.

I don't think this is totally true. The selling point of the cloud it's not only what you mentioned, maybe yes at the beginning.

Today, is more important the time to market. You can use existing services to build your application, there are customers that don't have access to a pool of data scientist to run their own ML training for example. These would rather benefit from high level services that help them to deliver functionalities just by using APIs.

> it becomes far more cost-effective and performance to just manage your own hardware and infrastructure.

This is very wrong too. First of all managing hardware is not just buying a server and placing it in your room. There are plenty of other business functions needed to run an effective datacenter. Procurement, hardware eng, sys admin, security (net, app, etc) and the list goes on.

You can do that if you reach a critical mass or if you are able to attract the right talents for that.

Big enterprise customers will settle for a suboptimal usage of their resources with limited capabilities to expand further or settle for one size fits all. You need a DB? We bought 15mil on Oracle licenses sorry about that, your no SQL needs or your vector search would be better make it work there.

As I said earlier, eventually this affects the time to market. Which is a far more important measure than cost for business. Especially thanks to the agility. If it doesn't work they tear it down, you don't need to keep paying for it.

I prefer GCP over some other providers as they don’t ask for a copy of passport. Digital Ocean or OVH is a no go because of that

Never had to do that with Digital Ocean nor OVH

I think they started asking for those a couple of years ago.

Guessing that still doesn’t cover China? GCP insists on billing me a cent every month for a handful of megs going there

Also no simple/reliable way to block egress to unwanted expensive regions where I have no users, especially with things like cloud run.

I’m pretty happy with ionos for my vrirtusl machines.

Also Cloudflare R2 for object storage - zero egress fees.

So basically up to $22 saved?

Don’t fall for it.

Reply on news.ycombinator.com