Google Compute Engine is now Generally Available | Hacker News Reader

Google Compute Engine is now Generally Available

googlecloudplatform.blogspot.com

92 pointsaryann29 comments

One salient new feature:

Transparent maintenance with live migration and automatic restart

At Google, we have found that regular maintenance of hardware and software infrastructure is critical to operating with a high level of reliability, security and performance. We’re introducing transparent maintenance that combines software and data center innovations with live migration technology to perform proactive maintenance while your virtual machines keep running. You now get all the benefits of regular updates and proactive maintenance without the downtime and reboots typically required. Furthermore, in the event of a failure, we automatically restart your VMs and get them back online in minutes. We’ve already rolled out this feature to our US zones, with others to follow in the coming months.

I'm curious about how regular hardware maintenance improves reliability. Are you just cleaning out dust? Or do you do periodic stress tests to catch failures early?

HW does fail over time and needs to be replaced. This involves shutdown of physical infrastructure at different levels.

regular maintenance helps with reliability because we do a lot of work proactively with power, hvac, switches, etc. which reduces likelihood of failure. Being able to perform maintenance in general is generally good for the stability of our system.

-Brian

People looking to test this out might want to try apply for the starterpack (https://cloud.google.com/developers/starterpack/ promo code brdo-in). I know a few people with pretty vague ideas that still got accepted, although I'm not sure if they'll be more selective now that the service is generally available.

Now you can run any out-of-the-box Linux distribution as well as any kernel or software you like. We’re also announcing support for SUSE and Red Hat Enterprise Linux (in Limited Preview) and FreeBSD.

====

Finally! I'm going to try and make TinyCore run: http://en.wikipedia.org/wiki/Tiny_Core_Linux

10MB images, 28MB of RAM, should boot in under 10 seconds.

Before we get into how terrible Google customer support is in general... has anyone actually tried their phone support for this service and can weigh in on it?

support is fantastic, we have a large cluster on GCE. I don't know how support will scale with general availability though.

You've called them on the phone? What were the incident(s) / support needs like on your part?

You know it seems very suspicious to create a new account to say this.

Especially since the common wisdom (and from my personal experience as well) is that Google's support is woeful.

I know that's the historical wisdom but we are working very and investing heavily to make sure that isn't true in general, and specifically for Cloud Platform. We've scaled our support team dramatically to support the influx of paid support with GA and are working hand-in-hand with lots of customers everyday. It's obviously never perfect, but I can assure you it's important to us and we are doing everything we can to make the experience with support as good as we can.

-Brian

Perhaps, and I'm not trying to be facetious, a detailed blog post regarding firstly, that this (customer service) has been an issue in the past, and secondly, the measures that have been taken ("We've scaled our support team dramatically") to ameliorate the support problem.

I see Google developing a bad reputation for customer service (on HN at least. See https://news.ycombinator.com/item?id=6837249 for the most recent example) and some honest recognition of the problem could go a long way to restoring this.

I apologize if you aren't the correct person to direct this to; perhaps you can divert it to the correct channel.

We are having amusing problems with appengine. For example they provide no way to limit the number of instances they decide to start. If for example you run out of datastore i/o capacity (it doesn't scale with how much you pay) they will quite happily start more instances even though they won't help in any way, and actually make things worse.

We switched to premier which supposedly gets us all sorts of extra stuff, but for some bizarre reason will no longer show your daily charge breakdowns.

Roger, sorry to hear this. Can you ping me at bgold at google.com so I can make sure you are getting the right help? Thanks!

There's no way to limit the number of instances that are started, but there is a way to limit the number of idle instances that you are paying for.

It is the former that matters to us. Essentially our code receives an HTTP request with payload, does sanity checks and formatting, and then a datastore put. This is not CPU bound - it is idle wait for the datastore to finish.

When you run out of datastore capacity, appengine will start more instances because it sees the lengthening processing times, but that just makes things worse. That you cannot stop.

Our load comes from mobile clients and no humans are involved. If they get an error they try again later. They don't care how long the server side takes to respond.

It is insane that you can't limit the number of instances. appengine would create hundreds at some points, which was very lucrative for Google (our daily fees went from ~$15 to $380 at one point).

I really look forward to a price / performance comparison between GCE and AWS. Persistent disk performance and no charge for IO looks really nice. It'll be great to see some real world numbers and benchmarks. Some key metrics I'm looking for are disk performance, CPU performance (how fast a single core is rated relative to amazon's ECU) and network performance!

There are some initial benchmarks here: http://serverbear.com/10151/google-compute-engine

The guys at Scalr were working on some benchmarking AFAIK. It's not up yet but I expect it will be soon-ish. http://blog.scalr.com/

This strikes me as an odd time to release the platform. My impression is that most companies avoid major infrastructure changes during the holiday season to maintain uptime for the surge in revenue. Even for B2B companies, it is the final month of the quarter. I would think that early January would be the best time launch and promote a new service that requires major infrastructure changes by clients.

You'd be surprised, for a lot of industries -- and for select departments of others -- the holidays are a source of relative slowdown. I've done more than one migration, roll-out, or refresh over the holidays, since there's usually less users to concurrently support (and thus downtime is less disruptive).

There are probably a fair amount of engineers on holiday vacation that might tinker with the platform for personal projects.

Amazon EC2 vs Google CE - Fight.

Does anyone just use EC2/CE though ? Surely it is about the entire platform.

Yes, lots of our customers do deploy to EC2 just to have the VM management facilities.

Conceptual overview of Compute Engine: https://www.youtube.com/watch?v=LqoXz-W3R80

No Ubuntu?

now i'm confused between app engine and compute engine :(

Reply on news.ycombinator.com