Hosting Options, should we stay physical or go cloud

Forums › IT Pro and developers › Hosting Options, should we stay physical or go cloud

maffey

43 posts

Geek
+1 received by user: 20

ID Verified

Lifetime subscriber

#296410 15-Jun-2022 11:41

We currently have 1/2 a Rack of physical hosting in an NZ data centre. There is a mix of colo, dedicated and shared servers in this physical space. This arrangement has worked great since 2015, with nearly Zero network downtime and only 5 or so incidents where servers have needed unscheduled physical reboots. Really good track record.

That said, our physical servers are getting old and will want replacing so I have been thinking about our hosting arrangement and if it is time to go cloud based.

If we stay with the physical hosting option, we will likely get new servers with more Cores, RAM etc and virtualise them using Proxmox.

These days, it seams like every man and his dog have moved from physical hosting to AWS or Azure.

We are looking at using AWS/Azure or one of the second Tier providers with a presence in Australasia like Vultr or Linode.

It seams cloud hosting options are cheap enough for small sites and applications, but get expensive quickly once you need some serious processing power and RAM. I can easily see our monthly costs doubling by moving to the cloud.

I like the idea of not being responsible for the physical hardware. Playing with hardware is fun, fixing broken hardware in the middle of the night, or finding someone to do it for you when you are away is not fun. We have the skills to manage hardware, but we are not experts at it, our focus is on web / application development.

Any thoughts, experiences or in sites welcome.

1 | 2

3689 posts

Uber Geek
+1 received by user: 2174

Subscriber

#2927826 15-Jun-2022 11:45

Hybrid is probably the way to go.

Put in the cloud what makes sense to be in the cloud, keep on-prem what makes sense to be on-prem.

maffey

43 posts

Geek
+1 received by user: 20

ID Verified

Lifetime subscriber

#2927828 15-Jun-2022 11:46

I am leaning that way, put live sites in the cloud, leave DEV, Staging, warm standby, backups in the Data Centre

amanzi

Amanzi
1354 posts

Uber Geek
+1 received by user: 331

ID Verified

Trusted

Lifetime subscriber

#2927831 15-Jun-2022 11:52

This is a deep and complex topic, and many will have different opinions on it. In my opinion... your last bit summed it up nicely: "our focus is on web / application development" So you should be focusing on that and leaving the hardware and OS management to others. And if you follow that logic, you should really be considering platform as a service or serverless computing so that you don't need to worry about deploying VMs or patching OS-level components. It's all about where you want to focus your energy.

gbwelly

1263 posts

Uber Geek
+1 received by user: 776

#2927834 15-Jun-2022 11:58

amanzi:

you should really be considering platform as a service or serverless computing so that you don't need to worry about deploying VMs or patching OS-level components.

100% agree. IaaS has all the annoyance of on-prem, just in another datacentre. SaaS and PaaS is where it's at.

darylblake

1172 posts

Uber Geek
+1 received by user: 410

Trusted

#2927838 15-Jun-2022 12:05

If you have resource hungry applications, e.g. require LOTS of RAM/CPU etc then they will probably be cheaper to run on your own hardware.

If, however you don't require huge amounts of resources, then in my mind cloud is probably a better option. I run a tech startup, and for the work we do cloud is better, because we mainly run web based apps, small databases and our email is done with Office 365.

Hybrid set-up is generally a viable option if you need a bit of both. Or if you are in some kind of transitional state. But if you want to do this properly with services like Azure ExpressRoute or AWS DirectConnect, then this can add additional costs.

Do go through the exercise to see if Cloud is a more viable option. Or continue to use co-location in a DC. Its worth re-assessing now.. but if something ain't broke or is working well... you may not need to fix it..

We use EKS, Some EC2, Amazon RDS and S3. Route 53 and SQS/SES. But we are a newish company and do not have lots of legacy systems/platforms.

gzt

18672 posts

Uber Geek
+1 received by user: 7807

Lifetime subscriber

#2927845 15-Jun-2022 12:20

What tasks do you have that require processing power?

GPU is expensive. Processing power seems reasonable.

Dyson

Shop now for Dyson appliances (affiliate link).

maffey

43 posts

Geek
+1 received by user: 20

ID Verified

Lifetime subscriber

#2927852 15-Jun-2022 12:30

gzt: What tasks do you have that require processing power?

GPU is expensive. Processing power seems reasonable.

I don't think any of our apps require a heap of processing power. It is mostly just a mindset change. At the moment, if a customers web application is getting a bit greedy I pop it on its own physical server with 8+ cores and 64+GB RAM, 2TB raid 1 SSD, problem goes away at the cost of about $1500 of hardware (second hand servers, new SSD). To rent that kind of horse power from a cloud provider, I am looking at $800 a month.

So I need to be a lot more careful with the sizing thing. But I guess with the cloud guys, if it is too slow, press the button to make it faster.

maffey

43 posts

Geek
+1 received by user: 20

ID Verified

Lifetime subscriber

#2927901 15-Jun-2022 12:34

amanzi:

This is a deep and complex topic, and many will have different opinions on it. In my opinion... your last bit summed it up nicely: "our focus is on web / application development" So you should be focusing on that and leaving the hardware and OS management to others. And if you follow that logic, you should really be considering platform as a service or serverless computing so that you don't need to worry about deploying VMs or patching OS-level components. It's all about where you want to focus your energy.

I believe Function as a Service is where we are heading. We are a PHP shop, only google seam to be offing FAAS with native PHP support. Azure and AWS are not native PHP, Digital Ocean are not in Australia/New Zealand.

Once FaaS is a bit more widespread, and supporting PHP, I have no doubt we will migrate in this direction.

darylblake

1172 posts

Uber Geek
+1 received by user: 410

Trusted

#2927946 15-Jun-2022 14:06

maffey:

gzt: What tasks do you have that require processing power?

GPU is expensive. Processing power seems reasonable.

I don't think any of our apps require a heap of processing power. It is mostly just a mindset change. At the moment, if a customers web application is getting a bit greedy I pop it on its own physical server with 8+ cores and 64+GB RAM, 2TB raid 1 SSD, problem goes away at the cost of about $1500 of hardware (second hand servers, new SSD). To rent that kind of horse power from a cloud provider, I am looking at $800 a month.

So I need to be a lot more careful with the sizing thing. But I guess with the cloud guys, if it is too slow, press the button to make it faster.

It sounds like the application is fairly in-efficent. How could you need that much processing power for a PHP application. There must be a huge number of concurrent users, or the processing could be cleared up a lot. I am fairly experienced with PHP. Either way with the cloud you can take advantages of the other tools the platform has to offer.

I would most certainly look at streamlining the application.

timmmay

20857 posts

Uber Geek
+1 received by user: 5349

Trusted

Lifetime subscriber

#2927969 15-Jun-2022 14:31

Many physical resources are oversized to deal with peaks and growth. In cloud migrations I often find we can get away with servers with significantly less resources, particularly for servers that can be horizontally scaled. Reservations can also reduce the cost significantly.

Cloud will need a whole new skill-set, including a firm or person with good qualifications and experience, or upskilling your team. Don't underestimate the cost of a cloud migration. Once you're there though you can start looking at cost optimisations, serverless, using PAAS etc.

I think it's worth considering cloud, but get someone experienced to help you with the monthly cost and migration calculations. By experienced, I mean someone who has done at least a few cloud migrations, and has certification with the cloud in question.

hio77

'That VDSL Cat'
13036 posts

Uber Geek
+1 received by user: 3896

ID Verified

Trusted

Lizard Networks

Subscriber

#2931460 18-Jun-2022 18:30

I'm sure I recall having a discussion about something like this recently :)

Alot of the advice here is pretty bang on, there will always be a use for bare metal in the hosting market.

It's all about where that break even pro and cons sit for yku, hybird offerings were you mix the two can often be an excellent option.

Recently moved a customer between datacentres with near zero downtime for better physical attendance in case of a fault, also going from multiple dedicated servers to cloud for their front-end end and Dedicated upgrade for their archival & big data side.

I forsee doing this sort of thing for plenty more folk as hybird options become far more avalible. Azure and aws turning up in nz will be the next big step where hybird compute options become fully a nobrainer.

#include <std_disclaimer>

Any comments made are personal opinion and do not reflect directly on the position my current or past employers may have.

Lego

Shop now for Lego sets and other gifts (affiliate link).

Tinkerisk

4798 posts

Uber Geek
+1 received by user: 3660

#2931478 18-Jun-2022 19:49

Hybrid is the way to go :-)

- NET: FTTH & VDSL, OPNsense, 10G backbone, GWN APs
- SRV: 12 RU HA server cluster, 0.1 PB storage on premise
- IoT: thread, zigbee, tasmota, BidCoS, LoRa, WX suite, IR
- 3D: two 3D printers, 3D scanner, CNC router, laser cutter

timmmay

20857 posts

Uber Geek
+1 received by user: 5349

Trusted

Lifetime subscriber

#2931492 18-Jun-2022 21:43

There's a place for hybrid cloud. Hybrid adds latency, inside a DC tends to be sub-ms but AU to NZ 35ms or so, not a huge deal for a single request but if you need many requests it can cause issues. Hybrid doesn't tend to be as resilient as cloud using multi AZ/region features. But hardware can be cheap, cloud can be expensive, so the risks may be worth taking to save significant money - it depends on the SLA / use case.

zaptor

745 posts

Ultimate Geek
+1 received by user: 39

#2931669 19-Jun-2022 13:50

maffey:

This arrangement has worked great since 2015, with nearly Zero network downtime and only 5 or so incidents where servers have needed unscheduled physical reboots. Really good track record.

Not bad for commodity hardware. Be interesting to know whether the unscheduled downtime incidents were hardware or software related.
However, mainframes are an entirely different league of reliability and redundancy (and I/O capacity).

maffey:

These days, it seams like every man and his dog have moved from physical hosting to AWS or Azure.

Indeed, it's like an insidious virus...

zaptor

745 posts

Ultimate Geek
+1 received by user: 39

#2931670 19-Jun-2022 13:50

maffey:

I like the idea of not being responsible for the physical hardware. Playing with hardware is fun, fixing broken hardware in the middle of the night, or finding someone to do it for you when you are away is not fun. We have the skills to manage hardware, but we are not experts at it, our focus is on web / application development.

Any thoughts, experiences or in sites welcome.

I don't know the fundamental architecture your business operates on (you didn't really say), so it's hard to make much of an informed suggestion.

However, sounds like you're a web shop, with current servers running in-country. So, you're effectively running in "a cloud", just without all the bells and whistles (services and redundant infrastructure) one would get with AWS or Azure.

If most of your clientele is NZ, then moving to Sydney will come at the cost of network latency, and being at the mercy of whatever arbitrary load-balancing is employed on the longer route.

Just bear in mind, the law of physics is a real thing. 1ms (millisecond) = 1,000,000ns (1 million nanoseconds), which is ages to a modern CPU.

1 | 2