22 comments

[ 3.0 ms ] story [ 64.9 ms ] thread
Besides the PR-approved statements here and there that don't add much to the meat of the topic, one should admire AWS' ability to innovate and add new features and services at this incredible pace.

Right now, in 2019, I'd say that most IT organizations can benefit from "some" use of AWS, at least, and that for certain use cases going close to 100% on AWS might also be beneficial, when you take into account the "human" costs.

The amount of R&D and investments that AWS does makes their products more and more compelling, comparatively to other products/services in the market (e.g. private servers).

I just wish that there would be less cool-aid drinking and more practical considerations when comparing prices between public and private cloud. I guess it will come over time.

(disclaimer: worked at AWS from 2008 to 2014).

Azure is steaming ahead at a breakneck pace, too. Everyone wants to capture the cloud. The amount of funding going into it is incredible.
100gbps is a bit more than a PR approved statement! It wasn’t much too long ago that AWS guaranteed any kind of bandwidth beyond a gig.
I was referring to the first two paragraphs of "customer obsession", which doesn't add anything substantial to the announcement.
My apologies if that came off as criticism when it was intended to be violent agreement.
No need to apologize!

Also, careful with guaranteed - AWS has his special wording about SLAs and guarantees, e.g. S3 "designed" for 11 9s of durability, which sounds like a guarantee but it isn't.

Disclosure: I work at AWS

Some organizations set high SLAs and play the odds that they will meet them, figuring that it's worth it in marketing dollars if they have to pay out.

For an AWS SLA like Route53's 100% Availability [1], the service is fundamentally engineered in a way that the SLA will be met. The S3 SLA can be found at [2]

[1] https://aws.amazon.com/route53/sla/

[2] https://aws.amazon.com/s3/sla/

"the service is fundamentally engineered in a way that the SLA will be met"

So... with the recent Route53 outage, do you still believe that statement to be true?

Jeff - are the disks slower (but bigger) than the i3 versions?

It looks like the IOPS and throughput are lower at the top end (looking at the two release pages).

You see something greater than 2-million IOPS?
Yes - the announcement for the older i3 instances (https://aws.amazon.com/blogs/aws/now-available-i3-instances-...) claims 3.3M IOPS (presumably on the largest size, with 15.2TB) So it looks to me like these are larger, slower drives?
It used to be 3.3M IOPS on 8 SSDs and now it's 2M on 8 SSDs but significantly cheaper; I wouldn't be surprised if they've replaced MLC with TLC.
8 TLC drives could easily hit 3.3M IOPS before... i3 have 8 1.9TB NVMe SSDs max, right? A 2TB P4600(TLC NAND) has 610,000 random read IOPS max, so that's 4.8M IOPS right there. A 7.68TB P4320 has 427,000 random read IOPS max, so that's 3.42M IOPS right there over 8 drives... Maybe they went from TLC to QLC with these ones?
I am supposedly on a stay-cation this week. I'll ask the team to check this thread.
I3en drives have lower peak IOPS compared to the ones on I3 but storage latency is 30-50% lower on I3en. Also, it is 50% cheaper than I3 on per GB basis.
Can you clarify what 'e' signifies and how it differs from 'd'?
e for expanded capacity (i.e. 60TB), similar to X1e which had expanded memory capacity. n for network-optimized just like C5n and P3dn
Curious that they do not mention _write_ IOPS.
Someone's gotta benchmark these and let us know