EC2 instance

EC2 is an Amazon flavor of virtualized server. You pay Amazon money and they will create you a virtual server in their datacenter. Can choose how much CPU, memory, and storage you need and then you pay per hour + a certain amount of bandwidth in and out. You can do just about anything you want with that server, as long as it's legal.
Edit: As an example. I have two tiny servers hosted in a cloud environment that run 24/7/365. I pay $335 every month. Each machine has two vCPUs, 8GB of ram, 2400 iops.
 
Last edited:
its a VM hosted within AWS' infrastructure - simple as that. Biggest difference would be the vcpu & memory allocation, as you cannot configure them separately (you can with storage/ebs) - think t-shirt sizes basically.
 
EC2 is an Amazon flavor of virtualized server. You pay Amazon money and they will create you a virtual server in their datacenter. Can choose how much CPU, memory, and storage you need and then you pay per hour + a certain amount of bandwidth in and out. You can do just about anything you want with that server, as long as it's legal.
Edit: As an example. I have two tiny servers hosted in a cloud environment that run 24/7/365. I pay $335 every month. Each machine has two vCPUs, 8GB of ram, 2400 iops.
Do your instances have to be up all time?
Are those reserved instances?
They’re not burstable right?

2400 iops is magnetic, low use, I’m assuming your data charges aren’t that high.

Are you paying enterprise licensing monthly?
 
Do your instances have to be up all time?
Are those reserved instances?
They’re not burstable right?

2400 iops is magnetic, low use, I’m assuming your data charges aren’t that high.

Are you paying enterprise licensing monthly?
They must run 24/7. Not reserved or burstable that I'm aware of. Very low IOPS and storage requirements. Don't know what enterprise licensing or reserved instance means.
 
Reserved just means that you commit to use/pay for that instance for a specific amount of time (1-3 years). It doesn't matter if its on or not, you pay for the entirety of the commitment. In return, you get significant discount (up to 70%). Useful for 24/7 workloads (like database workloads).
Enterprise licensing - just think of enterprise software companies (microsoft, oracle, sap...), and you're paying to use their software through agreements, typically and are bundled with support plans. If you're using MS office at work, then you have an ent agreement with MS most likely - for example.
 
The fees for an instance being on but not doing anything are really low. You only get charged as you start using compute and/or bandwidth.
 
The fees for an instance being on but not doing anything are really low. You only get charged as you start using compute and/or bandwidth.
Do you mean stopped?

If it's running, you're paying for compute - you can very easily generate hundreds of dollars a month in costs from a single instance
 
Do you mean stopped?

If it's running, you're paying for compute - you can very easily generate hundreds of dollars a month in costs from a single instance
It can be running but basically using no compute. Depends on what it’s doing.
 
It being idle vs busy will have no bearing on cost. As long as it's in the running state, you're eating the cost of the instance.
 
It being idle vs busy will have no bearing on cost. As long as it's in the running state, you're eating the cost of the instance.
The cost of the instance is almost nothing though. It’s single digit $ for some EC2’s I run that aren’t moving any traffic or doing much compute.
 
I'm glad you're getting away with some burstable micro instances but we're thousands of times that.
 
It being idle vs busy will have no bearing on cost. As long as it's in the running state, you're eating the cost of the instance.
I was about to say….there’s a reason to snapshot and terminate instances after they hit known .01 resource use for x amount of time.
I’ve gotten into enough fights with Application owners about why their sql DBs are snapshot after 4-6hrs daily scheduled use.
Push to s3 or Kafka for redshift consumption.
Dbs we’re just up eating $.
I’m still staying to get their guys to understand and pick a worker strategy to move them off “vm does this, application does that”.
It’s like everyone clings to an understanding of LAMP architecture, but I still have to deal with the realities of stateless architecture bc there so much more sys ops thought they get really uncomfortable contemplating.

Reason why I asked about reserved instances is that you can take your logs and renegotiate with your enterprise salesperson.
We do that on GCP and azure as well.

Licensing I was referring to is oracle, sap, etc.
OS licenses, not so much.
Often you’ll see orgs orchestrate those workflows if they’re not 24/7.
If they are then there’s another reason with that instance depth to negotiate license costs directly with the vendor.
Yeah, byo license bind then swap can be a pain but if your org was paying full pop retail for say 100 VMware clusters that’s a way to save costs.

Short of picking apart your applications to find where you could save with either hosted resources or remapping your applications to eliminate redundancy you got me.

The low iops volumes and low transmission isn’t an obvious one bc that appears to revolve around cpu consumption.
Resizing ala Chick fil a and their famous 9000 prod k8s clusters sounds like a replatforming exercise.
Walmarts thing wasn’t something read deep into but there was another repackage of a lot of cpu consuming CRUD.
 
They must run 24/7. Not reserved or burstable that I'm aware of. Very low IOPS and storage requirements. Don't know what enterprise licensing or reserved instance means.
Yeah dude. You need to sign up for reserved pricing for stuff you know is up all the time.
 
Yeah dude. You need to sign up for reserved pricing for stuff you know is up all the time.
100%, and he should take his logs and negotiate the rate lower than whatever the menu rate for his region, family, size.

It’s also worth analyzing the application for instance fit to type.

It’s pretty rare that everything is up and chugging.

I’ve seen orgs resisting running their end of month reporting with Spot commodity instances set to 3 cents/hr

They were just used to having Sage up doing nothing bc it made them feel good.

You can use functions at 1mil invocations a month for simple internal rigging for free instead of Splunk nodes littering the estate.

I digress, basic workload eval like we were doing a Colo refresh is #1

500k rows and hour vs 850k rows an hour basics.

Unless this org is ingesting batches at regular intervals, I bet there’s room to better orchestrate the instances.

Depending on what’s running routing logic, further reduce instance useage offloading to api gateway, lbs, etc

If there’s worker processes that can be off loaded, do it.

There’s just not enough application insight to make suggestions.
 
Back
Top