TL;DR -- I urge all readers to take this post with a grain/boulder of salt: The author is anonymous, prone to hyperbole and error, and makes multiple unverified claims. I'm sure he's a nice guy/girl and if we met in real life I'd be happy to converse over a beer -- but I find the language and misinformation in the post overly polemic and disingenuous. If this stuff interests you guys tune into the reinvent livestream next week: https://reinvent.awsevents.com/live-streaming/
All numbers are from EU-WEST-1 (Does anyone know why GCE doesn't have a eu-west-1a? only b,c,d? I'd be curious to know the story there... not trolling -- just curious.)
Graph 2: c4.large has 2 CPUs... n1-standard-1 has 1... comparison on price is strange? Your point below about optimizing for manageability doesn't really make sense to me as manageability would stem from config/deploy/etc. -- not from instance type? Perhaps your language isn't clear. You're anonymous but I'm guessing english is a second language for you (judging from sentences in post) so it's possible we're missing your point there. Could you clarify please?
>"an ancient virtualization technology" source? details? KVM came out in 2007. Xen came out in 2003. I don't consider either of those dates particularly ancient but whatever it's your post, use the language you want.
Graph 3: A c4.4xlarge has 16 CPUs, and 30 gigs of RAM. An n1-highcpu-4 has 4 CPUs and <4 gigs of RAM. Comparing the two on price is disingenuous. Your claims that these are the "production instances" don't make sense to me.
>Network Heavy: So this is a test between two t2.micros in different AZs: https://s3.amazonaws.com/ranman-code/2016-11-20+00.58.23.gif
It shows 1gbps... I ran it literally a minute ago...
So that's 1gbit... Do you mean inbound from public internet or outbound to public internet? I created an n1-highcpu-4 and had it talk to a c4.xlarge at 1gpbs in (both in eu-west) so your claim that you need a c4.4xlarge for 1gbps seems dubious.
Test Complete. Summary Results: [ ID] Interval Transfer Bandwidth Retr [ 4] 0.00-60.00 sec 7.40 GBytes 1.06 Gbits/sec 1297597 sender [ 4] 0.00-60.00 sec 7.40 GBytes 1.06 Gbits/sec receiver
It actually seems like Google's network was the limiting factor there because when I ran a similar test on a 10gpbs instance I couldn't get faster than 1gpbs. Which is fine because that's what's advertised -- just pointing out that the need for a c4.4xlarge is wrong. On a c4.large I got 900mbps.
>C4/m4/r3 have a hard cap at 220 mbits
This one is just false. What am I missing here? Where did you get this info? From eu-west-1 to us-east-1 I get faster than that using public routes and t2.micros. Internally across availability zones I get much faster than that.
Local SSD #s: Use PIOPS volumes, tunable size, lower cost, can attach to any instance type. Also consider the numerous other instance types that offer ephemeral SSDs.
>It is quite flexible. For instance, we could recreate any instance from AWS on Google Cloud. I don't think you mean any instance... but ok sure!
>Amazon does everything wrong, and Google does everything right, A message by an employee from Amazon than Google, not directly relevant but still a good read.
^ I don't think you read the above post when listing it as a reference. The title is ironic. Steve Yegge quit Google. Famously... on stage... He also goes on to say that Google can't do platforms. I'm sure that's changed in the past few years though.
In the end I'm super excited about both AWS and GCP they both have awesome products. I encourage people to continue researching and writing posts around these topics. It's hard to not take some of the criticisms personally when you're passionate and invested in your work (at least it is for me). I've got a google employee DM-ing on twitter calling me "petulant child" among other things. We don't need that. It's not helpful for our companies and it's not helpful for our customers.
I'll encourage everyone to tune in to the reinvent livestream on the 29th, 30th, 1st: https://reinvent.awsevents.com/live-streaming/