đŸ’Ÿ Archived View for dioskouroi.xyz â€ș thread â€ș 29448494 captured on 2021-12-05 at 23:47:19. Gemini links have been rewritten to link to archived content

View Raw

More Information

-=-=-=-=-=-=-

New EC2 Instance Types of re:Invent 2021

Author: adriaandj

Score: 116

Comments: 49

Date: 2021-12-05 11:30:35

Web Link

________________________________________________________________________________

hectormalot wrote at 2021-12-05 13:38:13:

The specific designation of Intel with the ‘i’ suffix is interesting. I’m surprised how fast this went from “everything is Intel” to “only take intel if you need something like SAP”.

I remember the first AMD instances a few years back and the 10% price/performance was nice, but at the same time didn’t feel like it would convince enterprise users to switch. Now with Graviton2&3, they’re apparently at 30-40% advantage and any big user would be crazy not to explore moving workloads over.

I expect the Intel-AWS negotiations today are _very_ different vs 5 years ago. This new naming is just one of many signals of a further downhill trend to Intel, and it’s likely to gain momentum in the next few years as the experience and ecosystem around ARM grows.

twoodfin wrote at 2021-12-05 15:06:48:

I’ve been curious for a while: Has anyone done the envelope math to estimate how much of the Graviton price advantage offered by AWS is a reflection of true price/performance vs. AWS taking a short term margin hit to encourage growth of a CPU platform they control?

ksec wrote at 2021-12-05 17:39:16:

> by AWS is a reflection of true price/performance

I assume you mean cost / performance as in Cost to amazon and not price as in price paid by consumers? Because Graviton is both faster and cheaper per vCPU. ( Amazon unit for a single thread )

>AWS taking a short term margin hit

It is hard because even if you can reliably extrapolate the cost of Graviton chip by wafer price. You will never know how they put the R&D into accounting. Do you have NRE cost in every single one those chips? ( which is what most people assume but not necessarily true ) Or do you move the whole R&D of Graviton into whole AWS R&D budget? And then you will have to calculate the power usage per workload to work out the power saving which gives you the TCO. Since 30%+ of Operating cost are going to energy usage. Being power efficient is also extremely important at the scale of AWS.

All in all I do not believe that Amazon is takin a short term margin hit ( and I dont understand where this narrative came from ), it is cheaper to produce, it uses less energy, and Amazon is pricing it as such, slightly cheaper than Intel and AMD's counterpart. Just like they did with the first generation of ARM. Graviton actually being faster per core just happens to be a bonus.

One has to remember. Moving to ARM / Graviton instances isn't free. You have to do your own testing and benchmarks to make sure it is indeed faster for your Apps. And that isn't always the case. You also have to make sure every part of your stack compile and works 100% on ARM. It just happens the cost saving is so gigantic lots of companies are spending time to move, which create a positive feedback loop in both open source and proprietary software to move towards ARM.

fomine3 wrote at 2021-12-06 02:32:49:

I'm not sure is it real but they obviously have incentive to sell Graviton instances cheaper than it actually achieved. By doing discount, they can encourage customers to use more Graviton instance (it reduces development cost per chip) and they'll get discount from Intel/AMD.

xmodem wrote at 2021-12-05 19:03:27:

> and I dont understand where this narrative came from

The paranoid might suggest it is coming from Intel's PR team, who might want AWS customers to believe that the savings might not be sustainable.

wmf wrote at 2021-12-05 16:52:49:

There's no way to know what price AWS pays for any processor. Having said that, it looks like ARM server processors do have an area efficiency advantage over Intel/AMD.

lizthegrey wrote at 2021-12-05 17:25:40:

Intel charges a much larger markup per chip than AWS can get from their own silicon + licensing ARM IP; the power consumption of Intel chips is also terrible (it's been cited that it's 60% lower power consumption for c7g instances compared to c6i instances, for instance, in the latest reinvent talk...) -- and power is what predominates operating cost day to day once the chip is fabbed/installed.

nielsole wrote at 2021-12-05 15:41:56:

Custom instructions / features that are only available on AWS silicon will improve their vendor lock-in.

And thus would warrant short term price dumping.

Future cost/performance improvements will allow them to raise their margin.

Kind of worried what that will mean for the portability of software long-term.

my123 wrote at 2021-12-05 15:48:54:

> Custom instructions / features that are only available on AWS silicon

If there are custom instructions added, AWS customers aren't aware of them. Feature set pretty much unchanged too. It's a non-effect on that front.

staticassertion wrote at 2021-12-05 19:09:34:

Are you 'locked in' to intel because of AVX? The cost of moving is recompiling your binary.

ant6n wrote at 2021-12-05 16:16:30:

Currently the cheapest spot instances with >30gb are

a1/32gb/16vcpu (early graviton) @ 8c and

M5zn/48gb/12vcpu (late Intel) @ 12c

I think the intel is better value. What are those graviton instances with better value (given that they are probably much slower than Intel)?

windexh8er wrote at 2021-12-05 16:27:50:

> I think the intel is better value. What are those graviton instances with better value (given that they are probably much slower than Intel)?

Intel instances are significantly more expensive when considering the cost for performance. In fact it's not even close [0].

[0]

https://www.anandtech.com/show/15578/cloud-clash-amazon-grav...

tyingq wrote at 2021-12-05 17:11:59:

The parent seemed to be specifically calling out Intel spot instances.

The key part being the cheapest available Graviton and Intel spot instances:

a1/32gb/16vcpu (early graviton) @ 8c and

versus:

M5zn/48gb/12vcpu (late Intel) @ 12c

Perhaps that will change as Graviton ages and availability changes.

ant6n wrote at 2021-12-05 18:09:15:

Mmh. Based on that Anandtech article, I guess the graviton2 cpus are actually similar in performance to intel ones. But still, my spot instance searches (especially also taking into account interruption frequencies), prices seem a bit of a wash. It doesn't look like Graviton is 40% cheaper. This may be different for on-demand or reserved pricing.

cldellow wrote at 2021-12-05 17:02:40:

I wrote a framework that does regexes of the Common Crawl. The Common Crawl is a series of captures of billions of pages of the WWW. Each member in the series has ~60 TB captures.

My framework runs a fairly simple pipeline: for each 1GB file, stream it from S3, decompress it, search it with a regex library, stream results back to S3.

A1 spot instances have consistently been cheaper for this task when measured on a $/GB scanned basis. Part of this might be that the framework doesn't need lots of RAM, so paying for more than 2GB per core on the non-A1 instances is just wasted expense for me.

orf wrote at 2021-12-05 23:26:34:

Wouldn’t a lambda be far cheaper?

cldellow wrote at 2021-12-06 00:10:11:

In my experience, no. Quite the opposite.

Lambdas can be cost effective for bursty loads, but if you can push sustained CPU utilization, as this example does, EC2 is cheaper.

For example, an ARM-based Lambda with 2 GB of RAM running for 1 hour is $0.09. An a1.medium is $0.025/hr on-demand or $0.0084/hr spot.

Part of this is that Lambda scales your CPU performance with your memory usage. So even if your code can run in 128 MB of RAM, there's no cost benefit to you if it's CPU-constrained, because it will run 16x slower than if you allocated it 2048 MB of RAM.

hedgehog wrote at 2021-12-05 17:46:19:

It seems unclear, the Intel has fewer cores (6 vs A1's 8) and probably less memory bandwidth per core, but then each core would be faster than the Graviton1. You'd need to benchmark your specific workload to find out.

cldellow wrote at 2021-12-05 18:03:19:

The a1.4xlarge has 16 cores, not 8. One of the sales pitches of the Graviton is that it has no SMT, so you get more predictable per-core performance.

hedgehog wrote at 2021-12-05 20:14:09:

Oops, yes, it's 16 cores vs 6.

scoopertrooper wrote at 2021-12-05 16:13:42:

> only take intel if you need something like SAP

Not even that any more.

From the article: "A key highlight is that M6a instances are SAP certified"

Plasmoid wrote at 2021-12-05 17:35:12:

That really surprising. I switch our workloads from m5 to m5a and there was a noticeable drop in performance and increase in latency.

I'm running in-house go services that consume from kinesis streams, so nothing strange like SAP.

jeffbee wrote at 2021-12-05 16:44:38:

The 6i instances are beating all amd and generally-available graviton instances in raw performance and performance-per-dollar on my workloads. I don’t think you’ll be well-served carrying this mental bias around. Better to actually try it and find out.

If you’re going to compare the paper performance of an unreleased machine to a 3-year-old Intel machine then yeah probably the former will win.

hectormalot wrote at 2021-12-05 19:05:37:

> I don’t think you’ll be well-served carrying this mental bias around.

I didn't mean to pick sides, or suggest anyone picks graviton by default. I meant to comment on how the communication around these new architectures by AWS has shifted, and what - I think - it signals about Intel's current and future market position with AWS.

For reference, my stuff runs on Heroku, I'm clearly not that concerned about price/performance ;)

JoshTriplett wrote at 2021-12-05 18:32:22:

> The 6i instances are beating all amd and generally-available graviton instances in raw performance and performance-per-dollar on my workloads.

I can confirm that:

Build time of a Linux 5.13.11 kernel with `make defconfig all`

c6i.32xlarge: 24.871411s

m6a.48xlarge: 27.622526s

Build time of a Linux 5.13.11 kernel with `make allmodconfig all`

c6i.32xlarge: 183.775255s

m6a.48xlarge: 205.326048s

128-way Intel seems to consistently beat 196-way AMD for software builds.

JoshTriplett wrote at 2021-12-06 02:45:58:

(Correction: 192-way AMD.)

mihaic wrote at 2021-12-05 14:24:40:

A bit tangential, but I've always hated how opaque AWS makes the process of picking instance types, since they hide the specs and prices behind layers of complexity, mostly to make it easy for developers to waste the company's budget.

Does anyone know any browser extension for instance to place information inline, such as specs and monthly/yearly pricing?

storyinmemo wrote at 2021-12-05 14:55:13:

Not inline, but

https://instances.vantage.sh/

is the way I've been looking at it for 5+ years.

belter wrote at 2021-12-05 15:15:42:

In case anybody is curious, this is done with the AWS Pricing API:

https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2...

fs111 wrote at 2021-12-05 18:59:20:

The pricing api is so weird. It basically downloads a CSV file and serves that up through code. Also none of the names in there match anything else in the AWS nomenclature. It talks about SKUs and in general feels like somebody leaked an internal API by accident.

The official docs literally tell you to manually search the file in a text editor.

slownews45 wrote at 2021-12-05 19:29:51:

"mostly to make it easy for developers to waste the company's budget."

Is this the case? I can't even find cloudflare's egress pricing online AT ALL. With AWS I can find pricing for anything I want.

Is there a specific instance price that you are trying to find and can't, you may be able to post here and someone can help you. The prices are on the AWS site, and also exposed via API's so there are a ton of third party services for them as well.

I wasn't aware that AWS had set up pricing "mostly to make it easy for developers to waste the company's budget". If you need help saving money AWS offers some services for this directly and again third parties can help you if you are struggling.

judge2020 wrote at 2021-12-05 19:36:28:

> cloudflare's egress pricing online AT ALL.

Cloudflare charges $0 for egress for the main service as long as it's for serving a website (eg. serving binaries or random other files isn't allowed if it's disproportionate to regular website visitors). for anything else, the pricing pages are usually close by - The Workers[0] site has pricing at the bottom, and R2 will also have $0 in egress pricing[1].

0:

https://cloudflare.com/workers

1:

https://blog.cloudflare.com/introducing-r2-object-storage/

slownews45 wrote at 2021-12-05 19:46:20:

The reason people need / use high egress (ie, far above the 1TB etc AWS gives per month) is for things like video sites, software updates / distribution etc.

I've been burned (too many times) by these "free" claims.

Will you absolutely guarantee (and pay me if needed) that I can host a large video website on cloudflare with no egress? This will be in PB not TB.

What I have found is that these sites with "free" are inevitably scams, that the types of loads that actually need high traffic egress result in that middle of the night email to squeeze you into some high cost plan (which does in fact have an egress price).

That said, if you will cover overage charges - I'll spin up a website on cloudflare. Do you understand bandwidth needs of video hosting sites? I'm sure porn sites are in 10-15PB per day range (ie, > 300PB/month).

PaywallBuster wrote at 2021-12-05 19:58:55:

it's not about the website content, it's about the content served from Cloudflare network.

Your website can be anything you want, but if CF bandwidth usage gonna be 99% for videos, you're not using it to serve the website, you're distributing content.

Which is not the use case they're servicing (with their "CDN" service)

They have a new service for that:

https://www.cloudflare.com/products/cloudflare-stream/

Which may be expensive too or difficult to compare (they bill per minute/instead of data)

For huge volumes, it's certainly better to find "specialized" providers or budget options like bunny.net

slownews45 wrote at 2021-12-05 20:25:05:

Umm, this is what is such a gimmick about these "free" services.

I run a video website on AWS, no one cares what I am doing, they charge me and done. I distribute software updates, no one complains, they charge me and done.

I run a video website on cloudflare, they say, oh, you are not running a website, you are distributing content, that's against the rules. Similarly if I host my software, my software updates, my management layer with firmware update service etc. Heads ups, when I visit these sites, for most normal people I am visiting a website.

Do folks not realize when they keep on repeating that egress is "free" on cloudflare that is a total lie? Hello, it's not free. You will get a call and they will squeeze you.

The streaming product is useful finally because we have at least a rough idea of price. Youtube avg video is 700MB/hr or so (mobile video is huge these days). That get's us to about 9 cents per GB.

Interestingly, this was I think the egress pricing they had for workers back when they were public about it.

AWS is 5 to 9 cents as well depending on usage so same range.

AWS Cloudfront is 2 cents to 9 cents depending on usage.

OK, so we are within the normal range of things now with egress pricing.

In my own experience with this type of player, you still get the call if you optimize for their pricing (ie, you run your 4K content through them, and your other content via another platform) because despite their claims bandwidth is "free" they are actually losing money on it, and if you use enough of this "free" stuff their marketing budget can't handle it and they just shut you down.

It's just not worth dealing with.

You'll know Cloudflare has a competitive CDN offering when they are willing to give you the pricing publicly of providing a CDN service that can handle any type of content. That is the proof point.

Cloudflare does have Argo which I think runs in that 5 to 9 cent range again per GB.

Back to the claim that AWS has hidden pricing (especially relative to cloudflare). I still find that ridiculous. Cloudlfare seems to be a heavily marketing based org with very hidden pricing (everything is call for a quote). Does anyone know what % of expense cloudflare has in marketing / selling? I'd imagine pretty damn high.

judge2020 wrote at 2021-12-05 21:01:01:

Egress pricing is free for certain content. It's not free for all content. If i'm a fast food joint and offer free meals to veterans, I can do that, even if some customers complain that i'm not running a soup kitchen where everyone eats free.

Cloudflare can offer free bandwidth to a whole host of sites and still have "a competitive CDN offering" without giving away free video hosting, arguably one of the most expensive use-cases to support.

slownews45 wrote at 2021-12-05 22:56:51:

Sure. Cloudflare can give away free services as a marketing effort. AWS gives away 1TB/month of cloudfront bandwidth, 1 million lambda calls per month, 1 million SNS notification per month, 25GB of DynamoDB capacity, 1 million Simple Queue queries per month and plenty more. So does google.

But what, again, is the pricing of cloudflare's "competitive CDN offering"? That seriously all I am asking for.

If you are going to go on about pricing (cloudflare does) have the courtesy to list yours publicly and don't have gag orders in your agreements.

The context of all of this is Amazon supposedly obfuscating their pricing to waste budgets. With cloudflare, you can't even FIND their pricing and will need to talk with a salesperson.

Anyways, for those stumbling around with this, minimum I've heard cloudflare doing is $24K per year on their paid stuff, and then they added Argo at around 10 cents per GB in the pay as you go offerings. Cloudfront is 2 to 9 cents per GB, google 2 to 10, fastly is around 8 to 12 cents, keycdn 4 cents etc. Not claiming AWS or Google are the cheapest, but pricing is pretty darn upfront.

Also have heard of folks getting busted by cloudflare if too many folks in asia/non-USA use the service, this seems far fetched, but AWS does charge differently by geo area so not sure if cloudflare has some rules like that tucked away.

stillicidious wrote at 2021-12-05 15:58:05:

It's by design, many features of EC2 are unrelated to the chip itself. For example burstable scheduling, availability of enclaves, IO and networking bandwidth and a variety hypervisor features are all tied to the instance family. These are often far more important to a buildout than the exact chip you're running on

big-malloc wrote at 2021-12-05 16:15:53:

I gave up on the official site and now use

https://ec2.shop

— really nice for doing a quick price comparison

slownews45 wrote at 2021-12-05 19:32:14:

Most of these third parties are using official API's to get pricing data, if there is something with a hidden price do ask and someone can probably help you figure it out. One reason AWS got popular is that it actually was historically much HARDER to get pricing from your internal IT team / vendors like IBM etc. So AWS wasn't the cheapest (by any means) but was pretty clear on cost. So it's interesting to hear that AWS is purposely hiding pricing and making price complicated to "waste company's budgets".

nijave wrote at 2021-12-05 16:12:35:

Types

https://aws.amazon.com/ec2/instance-types/

Prices

https://aws.amazon.com/ec2/pricing/on-demand/

Is there something missing from these? Admittedly they can be a bit hard to find sometimes but I think that's a problem with AWS docs in general

Griffinsauce wrote at 2021-12-05 15:37:42:

If you have the option, you can also vote with your wallet and go somewhere else.

I found the upfront pricing of DigitalOcean very refreshing after AWS.

mark_l_watson wrote at 2021-12-05 14:42:05:

I mainly use TensorFlow on AWS, with or without GPUs. I would like to read a reference for Gravitron, etc. support for TensorFlow.

unixhero wrote at 2021-12-05 16:49:49:

Watching reinvent on YouTube has been very profitable for my consulting offerings. Thanks AWS!

soheil wrote at 2021-12-05 16:32:14:

Gravitons haven't been discovered yet, if they do exist they're probably as difficult to find as their counterpart AWS instances.

gatestone wrote at 2021-12-05 16:29:56:

If you use the Amazon Kubernetes service or similar at Google or Azure (EKS, GKE, AKS) you can pretty much let them manage your VMs. With Terraform it’s a breeze, start with the defaults and try different setups.

systemvoltage wrote at 2021-12-05 16:51:12:

Googling “AWS EC2 benchmark” leads to nowhere. Does anyone have a good source of objective instance comparison between cloud providers? I went off one day and created instances on AWS, GCP and DO - and ran Geekbench5 (yes I know it’s not a professional benchmarking tool but better than nothing I suppose) and found large differences. DO was the slowest of all. Do most companies do their own due-diligence before committing large sums of money into AWS?

wilde wrote at 2021-12-05 21:55:26:

I’ve been using

https://www.vpsbenchmarks.com/

but haven’t done much in the way of independently confirming their rankings.

slownews45 wrote at 2021-12-05 19:34:41:

Digital Ocean is really serving a different market. The folks recommending them tend to be running smaller workloads (so DO can have relative lower perf per vCPU etc).

The best I've found - run the workload you actually plan on running. Even on AWS etc, the perf differences are getting pretty meaningful between instance families even at similar price points.