Multiple Dedicated Hardware for everything or Virtual Instances from one beefy server ?

[soamz]

As im building the business plan for what hardware to order to dell or super micro, I need some help, as Im getting mixed opinions.
Im going to offer Shared Hosting with HDD, Shared hosting with SSD, VPS Hosting with SSD, and Dedicated BareBones (when we get order, we will buy 1U server for each dedicated server client)

Some say, go for dedicated hardware for each type of hosting, means they said,


get 1 server like single Xeon E5 1585 with 32GB of RAM and you can host like 1500+ shared hosting HDD clients and shared hosting SSD clients easily.

get 1 server 1 server like dual Xeon E5 1585 with 96GB of RAM and you can host like 100+ shared hosting HDD clients and shared hosting SSD clients easily.

And get 1 big Supermicro with 24 bays for RAID backup and auto backups of both shared hosting server and VPS server both to act as SAN storage and emergency backup restore.
And some said,

get one beefy server, run Xen server in it and create 2 huge instances.
One instance use Cloud Linux and offer shared hosting from it.
And one instance use KVM+Openstack and offer VPS hosting from it.

And And get 1 big Supermicro with 24 bays for RAID backup and auto backups of both shared hosting server and VPS server both to act as SAN storage and emergency backup restore.
Now Im confused, which one to finalize.

I think,1st plan is better, as the fail rate would be much less, and we wont be depending on one single big server for everything, because if it fails, then whole business will go down until we replace it.

What do you think ?

Which is better ?

 

[T_Minus]

To me it would depend what 'type' of clients you are hosting. Your other thread mentioned 1000 existing clients so you probably know a bit about them I would never cram that many clients on one server especially an "all in one" setup that runs the web + db on itself.

Depending on budget I would setup something like a proxmox ve cluster with minimum of 4 nodes. Again, I don't know your client utilization but if you're planning on jamming 1500+ shared on a 4 Core system I can only assume most are small, static sites that mean nothing to your clients.

Server #, specs, and performance aside I wouldn't want to jam more than 100 clients on 1 server, especially if you have some type of uptime agreement, or your clients will call you if something goes down... sounds like a huge issue/headache that could lead to catastrophic failure of their site/business and potentially yours if they're relying on you.

 

[soamz]

Okay its basically 90% of them are wordpress websites clients, who has like 2-3 websites in their hosting accounts.

They hardly pay 5$ a month. So, I guess, a 4 core 1280v5 server can easily handle around 1000+ wordpress sites for sure.

 

[ttabbal]

I think T_Minus is on the right track. Even with low utilization clients, 1000 wordpress sites have DB lookups and such. And you have to consider downtime/maintenance. At some point, you need to do updates on the host OS. A cluster would let you do that without killing 1000 client sites. At $5/mo, I doubt there is a significant SLA, but still... I assume at some point you might want to target more expensive plans, and getting yourself a reputation for lots of downtime and performance problems is going to hurt you there.

As you are talking about an existing setup, what is the current hardware and utilization like? That seems like a good place to start to understand your needs.

 

[soamz]

Yes, thats the reason, Im not looking to go crazy on the hardware, as the 5$ clients will be the most and they also know that, it doesnt come with any kind of SLA.

So, Im looking to stack all those small clients in a server and for serious wordpress clients, I would put them in a SSD server may be.

 

[_alex]

i'd go for a cluster, what gives you freedom to service or upgrade single nodes without downtime.

ha-enabled vm's and/or a pair of loadbalancers in front of multiple application-servers could also be worth to think about.

also, going for galera-cluster + (asynchronous) replication to another site for the db's instead of all-in-one lamp setup is something to consider.

with 3+ nodes with either hyperconverged Storage (ceph, gluster, sheepdog) or an additional shared storage you should be fine.

also, how do you plan your strategy for backups/dr?
how long it would take to come back to operation from backups is a good thing to know in advance

alex