7 ms·

The Next Generation (I3en) of I/O-Optimized EC2 Instances

59 points by jeffbarr 7y ago

simonebrunozzi 7y ago
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).
- asdfman123 7y ago
  Azure is steaming ahead at a breakneck pace, too. Everyone wants to capture the cloud. The amount of funding going into it is incredible.
- mattrp 7y ago
  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.
  - simonebrunozzi 7y ago
    I was referring to the first two paragraphs of "customer obsession", which doesn't add anything substantial to the announcement.
    - mattrp 7y ago
      My apologies if that came off as criticism when it was intended to be violent agreement.
      - simonebrunozzi 7y ago
        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.
        _msw_ 7y ago
        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/ https://aws.amazon.com/route53/sla/ [2] https://aws.amazon.com/s3/sla/ https://aws.amazon.com/s3/sla/
        adreamingsoul 7y ago
        "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?
dhendo 7y ago
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).
- p0rkbelly 7y ago
  You see something greater than 2-million IOPS?
  - dhendo 7y ago
    Yes - the announcement for the older i3 instances (https://aws.amazon.com/blogs/aws/now-available-i3-instances-for-demanding-io-intensive-applications/ 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?
    - wmf 7y ago
      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.
      - Rafuino 7y ago
        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?
    - lfittl 7y ago
      This seems to be confirmed by the IOPS numbers in the documentation: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/storage-optimized-instances.html#i2-instances-diskperf https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/storage-...
- 7y ago
awspm84 7y ago
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.
- solidasparagus 7y ago
  Can you clarify what 'e' signifies and how it differs from 'd'?
  - AWS_F1 7y ago
    e for expanded capacity (i.e. 60TB), similar to X1e which had expanded memory capacity. n for network-optimized just like C5n and P3dn
makeshifthoop 7y ago
Jeff - Looks like there are some mistakes here https://aws.amazon.com/ec2/pricing/on-demand/ https://aws.amazon.com/ec2/pricing/on-demand/ i3en is showing "EBS only" under "Instance Storage"
foobiekr 7y ago
Curious that they do not mention _write_ IOPS.
- Rafuino 7y ago
  Someone's gotta benchmark these and let us know

59 points by jeffbarr 7y ago

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).

asdfman123 7y ago
Azure is steaming ahead at a breakneck pace, too. Everyone wants to capture the cloud. The amount of funding going into it is incredible.
mattrp 7y ago
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.
- simonebrunozzi 7y ago
  I was referring to the first two paragraphs of "customer obsession", which doesn't add anything substantial to the announcement.
  - mattrp 7y ago
    My apologies if that came off as criticism when it was intended to be violent agreement.
    - simonebrunozzi 7y ago
      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.
      - _msw_ 7y ago
        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/ https://aws.amazon.com/route53/sla/ [2] https://aws.amazon.com/s3/sla/ https://aws.amazon.com/s3/sla/
        adreamingsoul 7y ago
        "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?

dhendo 7y ago

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).

p0rkbelly 7y ago
You see something greater than 2-million IOPS?
- dhendo 7y ago
  Yes - the announcement for the older i3 instances (https://aws.amazon.com/blogs/aws/now-available-i3-instances-for-demanding-io-intensive-applications/ 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?
  - wmf 7y ago
    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.
    - Rafuino 7y ago
      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?
  - lfittl 7y ago
    This seems to be confirmed by the IOPS numbers in the documentation: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/storage-optimized-instances.html#i2-instances-diskperf https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/storage-...
7y ago

awspm84 7y ago

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.

solidasparagus 7y ago
Can you clarify what 'e' signifies and how it differs from 'd'?
- AWS_F1 7y ago
  e for expanded capacity (i.e. 60TB), similar to X1e which had expanded memory capacity. n for network-optimized just like C5n and P3dn

makeshifthoop 7y ago

Jeff - Looks like there are some mistakes here https://aws.amazon.com/ec2/pricing/on-demand/ https://aws.amazon.com/ec2/pricing/on-demand/ i3en is showing "EBS only" under "Instance Storage"

foobiekr 7y ago

Curious that they do not mention _write_ IOPS.

Rafuino 7y ago
Someone's gotta benchmark these and let us know