Bhaskar's picture

Hi Team,

We already have a subscription for "LAMP Stack - Web Stack (MySQL) powered by TurnKey GNU/Linux (HVM)" in our AWS account and I am raising this ticket on behalf of our customer. The issue that we are facing is that we are not able to launch the EC2 instance of types t3* and m5* series. Could you please help with this as we need to optimize the instances. Could you please consider this as the highest priority and help us.

Forum: 
Jeremy Davis's picture

AFAIK our Marketplace AMIs are theoretically ready to support t3 (and therefore I assume m5 too) but they're not yet well (enough) tested IMO. Unfortunately I got pulled off that task onto items deemed higher priority.

If you're willing and able to provide feedback and accept the possibility of hiccups, please respond ASAP and I'll aim to prioritise this over the next week or two?! Also a commitment to give us a positive (although obviously honest) review on AWS Marketplace would also be a big plus for us and would be warmly welcomed! ;)

Bhaskar's picture

Hi,

Is there any update on this. who can help us on this.

We are still not able to deploy the t3* and m5* series.

Jeremy Davis's picture

I'm still tied up with other stuff and haven't even looked at this yet.

I was (and still am) willing to prioritise this, but I didn't hear back from you?!

Let me repeat myself:

If you're willing and able to provide feedback and accept the possibility of hiccups, please respond ASAP [...]

If I do stop what I'm currently working on and start working on this, I can't be sure how long it will take, but probably best case scenario would a few days. It's also worth noting that there may be factors outside of my control (e.g. AWS MP have to approve the changes and I have no control over how long that might take).

Having said that, you seem like you are in a rush. I don't really want to drop what I'm doing now to focus on this and it ends up that you can't or won't wait the time that it will likely be before the new types are available to launch from AWS.

I look forward to your response and once I hear back, I'll try to keep you up to date with how things go.

Bhaskar's picture

It is not about the rush, and when you said that you have to have a look into that, we were expecting some update. It is fine if you take a week or two and it is fine if the upcoming rlease supports the required instance types, but we want to know whether current release that we are using supports the required types or not.

Jeremy Davis's picture

It appears that there has been some mutual misunderstanding. Apologies for my part in the poor communication. A friend of mine often says that the biggest issue with communication is that we assume that it's happened... :)

Anyway, let's reset and start again...


The AWSMP page that you link to, notes the server types/sizes which are currently allowed. As noted there (and your experience too), the current LAMP appliance does not allow t3 or m5 sizes/types.

To be crystal clear: it's not currently possible to launch t3 or m5 instances of that AMI!

There are some additional technical AMI requirements for running these newer instance sizes. These technical aspects have been addressed in theory and they should work fine under the newer types/sizes. The basic tests we ran all passed so I'm fairly confident that they'll work fine. However, I can't be 100% sure. That's because testing never moved beyond the basics (e.g. we just launched a few instances and did basic usage testing). It's possible that there may be edge case issues that require us to release a new version to address those.

As we didn't do any advanced testing, we never enabled use of the new instance types/sizes from the Marketplace. Experience suggests that if a new user encounters issues with a Marketplace AMI, they rarely report it. They usually just trash it and move on to another product. So not allowing use of the new instance types/sizes seemed like the safest option until we circle back to do further testing.

As I noted previously, I'm currently tied up with other stuff and I'm not sure how long that might take to get back to testing the newer AMI types/sizes and how long that testing might take. So until I do get back to that (which might be a while away), the only other way I could be convinced to enable the use of these instances sizes in AWSMP would be if a user committed to provide feedback on how they go.

So in an effort to reword my original post: if you are willing to commit to provide feedback on how the newer instance sizes go, I'd be happy to enable them. But I need you to provide that commitment first! In return, I'd be willing to commit to address any issues that may occur and at the very least seek to develop a workaround and/or create a new bugfixed AMI.

Does that make sense now?


To summarize; in relation to our AWSMP LAMP appliance:

  • t3/m5 are currently not allowed in AWSMP
  • theoretically they should meet the technical requirements and work fine
  • if you commit to providing feedback on the new sizes, I'm willing to do a quick double check that they work ok and ask AWS to enable t3/m5 instances sizes ASAP
OnePressTech's picture

Hi Jed, I am also upgrading servers and need T3/T3a enabled...please :-) I will commit to feedback on the new servers if you enable them. What technical issues might there be? Is the potential risk the T3a because it is ARM based?

Cheers,

Tim (Managing Director - OnePressTech)

Jeremy Davis's picture

From my understanding, they should either "just work" or not. It's been a while now since I looked at it, but IIRC most of the changes are in the network stack requirements. As I noted above, the work has (or at least should have) been done.

So TBH, whilst I don't anticipate any issues, as I noted above, we haven't done extensive "battle testing" on them.

So what I might do then, is enable t3/m5 instance sizes in the current LAMP image and await feedback. If need be, I can make tweaks and rebuild.

I'd also be really interested to hear your experience with t3 instances in general. Whilst on face value they do look like an improvement, I have heard a few reports that depending on your workload, t2 still generally provide superior "bang for buck". One report I read compared the 2 and for the particular workload they were using, overall t3 was ~10% more expensive for slightly less performance.

From what I've read, with the t2 instances each "vCPU" actually maps to a CPU core, whereas on a t3 instance, each "vCPU" actually only matches to a thread (i.e. Intel server CPUs use "hyperthreading" to provide 2 threads per CPU core). Obviously your results will depend on your workload (i.e. YMMV).

Re t3a instance sizes, AFAIK the significant difference is that they run on AMD chips, rather than Intel. As per Intel chips behind t3 instances, the AMD "EPYC" CPUs provide 2 threads per CPU core and as per t3, each "vCPU" is a thread, rather than a CPU core. The t3a Benchmarks that I've seen suggest that the performance of t3a instances is slightly lower (~10%) than t3, but OTOH from what I've seen, the price is about ~10% lower too. So roughly equivalent "bang for buck". My reading to date suggests that there are no new technical requirements for t3a, so assuming that is correct and the current images are fine on t3, then this should "just work" too...

We'll see I guess. I'll aim to get this done in the next couple of days.

OnePressTech's picture

Regarding T2 vs T3, my understanding is that they are equivalent in that T2 has 1VCPU (core) and T3 has 2VCPUs (2xthreads). I guess the question is whether T2 runs on a dual core or quad core CPU.

The following company analysed T2 vs T3 and found T3 was better than T2 for linux but T2 was better than T3 for Windows. https://www.densify.com/articles/aws-ec2-t3-burstable-instances

Regarding a comment the author made...

"T3 instances are powered by the Nitro system which means that they support network and EBS bursting. This also means that you must watch out for things like HVM virtualization exclusively (no Paravirtual) and the need to launch within a virtual private cloud (VPC) using an AMI that includes the elastic network adapter (ENA) driver."

...does the TKLX VM contain a new elastic network adapter (ENA) driver which is what you have not tested specifically?

Cheers,

Tim (Managing Director - OnePressTech)

Jeremy Davis's picture

Ah I missed that t3 has 2x "vCPU" (i.e. 2 threads/1 core) and t2 has 1x "vCPU" (i.e. 1 core/2 threads).

IIRC the benchmarks that I saw (that showed significantly better results for t2 and marginally worse results for t3a; both compared against t3) were done on Windows. So you may well be right. I should have dug a little deeper! :)

...does the TKLX VM contain a new elastic network adapter (ENA) driver which is what you have not tested specifically?

IIRC it's in the Linux kernel. I forget the details now, but I did need to make some tweaks to enable it to work as expected, but those changes should already be part of the current AMIs. When I implemented them, I did do some initial t3 tests and it seemed ok (i.e. boot LAMP, can I connect? Yes - pass!). I had wanted to do some more extended "battle testing" to ensure that there weren't any other issues that my simplistic testing missed.

I think that there were also changes re default mount points, but it's a bit fuzzy...

I'll redo the basic t3 testing with the current LAMP image and see how we go... Assuming that they go ok and the AWSMP API hasn't changed significantly (they have reported that it will be updated again soon which may break my update script) then hopefully we can get them enabled by week's end, early next week at the latest (fingers crossed).

OnePressTech's picture

Cheers Jed. Sounds like a plan...much appreciated :-)

Cheers,

Tim (Managing Director - OnePressTech)

OnePressTech's picture

Bump :-)

Cheers,

Tim (Managing Director - OnePressTech)

Jeremy Davis's picture

I'll get onto this within the next day or 2... :)

I'll aim to keep you posted via this thread.

Jeremy Davis's picture

So I launched a t3a.micro test instance of our LAMP appliance and all went well. So I issued a change request to AWSMP API but I got an error.

I see a likely workaround for the error, but reviewing the error bought another potential issue to my attention. It looks like I'll be having a call with AWSMP early next week, so that will delay things a little.

Sorry about that, but at least it's "in progress" now. So hopefully we should get appliances supporting t3/t3a/m5/m5a really soon.

OnePressTech's picture

It will be good to have access to those images. Much appreciated :-)

Cheers,

Tim (Managing Director - OnePressTech)

Add new comment