Taking The Jump (Leap!) to AWS: Performance, Costings and other things

Taking The Jump (Leap!) to AWS: Performance, Costings and other things

I appreciate that most companies now exist within the AWS or Azure clouds, and will be a whizz at scripting, loading balancing and billing. But for me, it’s a big move to take a server that I have run for five years, and which has run — mostly — well, and move to AWS.

While performance is a worry, especially as I have 16GB of server memory on my current server, and six cores, it is cost that it the major worry, especially as I have to pay for it myself. But it has supported my teaching for many years, and has been a platform for me to learn new methods. So, with my baremetal server now down to a secondary plexar boot, and a console login required to reboot my server, it needed to migrate to the public cloud. Also, my 1&1 server has a small 60GB disk for c:, but 1TB for d:. As it’s Windows, unfortunately, much of the software goes onto c: by default, so I just could do anything properly with it. I ended up creating soft links all over the place. I also wanted Visual Studio on it, but there’s no way it could fit on there, even with the 1TB disk. So, as the semester has now finished, I’m taking the jump (leap, in fact).

We build a whole virtualised infrastructure using ESXi, and it works well. Currently we support thousands of students, and with complex networked infrastrutures. And for my teaching, I’ve developed asecuritysite.com in order to support students. And, so, I’ve avoided AWS, as I could never understand the billing. For some reason, every month I received a hit on my credit card, and I never had time to dive into the details of this, and truly understand the bills. But my old 1&1 server is creaking, and it has problems booting, and I can’t really back it up properly. I also can’t ramp it up with improved hardware when there’s a test, and bring it down when times are lighter.

So, sometimes, you need to take a jump, and I’ve moved it all to AWS, and finally I understand the billing. I was — nievely- under the impression that when I stopped an instance, that the billing would stop. But, I was forgetting about the volumes and the snapshot, and where you pay for these, even if the instance is on or not. And, at that point I realised that it is not a great idea on adding a 500GB disk onto a Web server. The basic costs were quite simple:

• $0.05 per GB per month for a snapshot.
• $0.10 per GB per month of a volume.

And so I had old instances that I was prototyping, and each with a 500GB volume. Thus the cost for the volume was around $50 per month (!!!), and my snapshots were coming in at $25 per month. So you must watch when you create your instances, not to add too much disk space. I all made sense, if I deleted the volumes and snapshots I didn’t need, I could get the cost down below my 1&1 baremetal server, and have backups. And, so with a bit of trepitation, I clicked on disk management on my Windows server, and reduced my partition size to 250GB. I then added a new volume for 250GB, mounted it onto my server. Next it was a simple install of Macrium Reflect, and I mirrored by c: partition to the new partition. Once finished, I stopped the server, disconnect the original volume, and added the new one as /dev/sda1, and restarted, and magically my volume was 250GB. A quick snapshot of the new volume, and then delete all the old 500GB volumes. A saving of — at least $25 per month, and for 10 minutes of effort.

With the budget trimmed, I could see straight away the reducing in costs, as it’s all done on a per hour basis, so the trimming reduced the costs immediately. So, now I know my base line, I could pick my server, and while my baremetal server running a Xeon processor has 4 CPUs and 16GB, I found that the t3a.medium EC2 server could comfortably cope as fast with its 2 VPCUs and 4GB. Whether it will cope with large loads, is still to be seen, but I can now script my server to ramp up and down as required. Before exams, I move the server up and give it more CPUs and memory, and then after the exams, it can move down again. The hourly costs are:

• t3.medium $0.04 2VCPU, 4GB
• t3.large $0.08 2VCPU, 8GB
• t3.xlarge $0.16 4VCPU, 16GB

Again, Amazon has done a good job in costing these, as it’s double the cost whenever you need to move up. So for me, perhaps I need 4 VCPU and 16GB (as my baremetal server) when I’m running in term time, with Splunk machine learning being used.

And so for me, now, I understand my base costs for the volumes and snapshots, and can then budget for the required server hardware. But, for the management of the server, things are a whole lot simplier. There’s a few things to get running, but I should be able to say goodbye to baremetal, soon.

If you spot any problems, please say:

From my tests so far, 2 VCPU/4GB on EC2 beats a baremetal server with AMD Phenon II X6 with six cores, 2GHz clock, and 16GB of memory.

Please note, asecuritysite.com is free of adverts, and is not paid by an external income. It does not have a 24x7 support team and a whole lots of developers. It’s just me!