Servers + Storage - what would you spend £££ on?

[Paid_Peanuts]

In a theoretical world, if you had a budget of £50,000 to spend on refreshing your ageing server equipment (compute + storage), where would you put your money?

 

Say you needed to run 50 VM's approx. and had a 30TB requirement, and weren't bothered whether you ran VMware/Hyper-V what would you do?

Would you:

• Use vSAN or equivalent like: DataCore/StarWind?
  
• Use Microsoft Storage Spaces Direct S2D?
  
• Keep storage local to the hosts and run VMs locally and invest in a suitable backup solution in case of host or storage issue?
  
• Use 2/3/4/5 hosts?
  
• Would you buy a SAN, if so what kind of disks would you like to fill it with?
  
• If you were to buy a SAN what connection type would you choose? SAS/iSCSI/Fibre?
  

Edited February 5, 2018 by Paid_Peanuts

[dhicks]

In a theoretical world, if you had a budget of £50,000 to spend on refreshing your ageing server equipment (compute + storage), where would you put your money?

 

Assuming we kept local hardware instead of using AWS or similar: probably 3 servers, vSAN or equivalent, tiered storage in each server with a couple of SSDs backing 6 10TB SATA drives in RAID 1 (so, hopefully, SSD performance with cheaper SATA drive capacity). At least 96GB of RAM in each server (not quite sure where the current point of server-RAM-gets-really-expensive is).

[PotNoodleTech]

For £50k I'd go VSAN on 3 really good hosts & a Veeam Host.

For £30k I'd go Storage spaces on 3 good hosts and a Veeam Host.

For £20k I'd go HyperV 2 good hosts with local storage and a Veeam host.

 

If it were my money I would be frugal, grab 3 really good hosts and a veeam host second hand from something like serversource for about £20k inc all software, then pocket the remaining £30k and spend it on something useful, like 3 x IT suites.

[nicholab]

I would go hyper converged with Datacore.

[Paid_Peanuts]

For £50k I'd go VSAN on 3 really good hosts & a Veeam Host.

 

Problem is because of the storage efficiency of using vSAN or DataCore, the cost of the performance disks (SSD) cripple the price.

Sure, you can buy plenty of 4/6/8TB drives for the capacity, but the costs of the SSD to maintain a 20& flash ratio from the likes of HP/Dell/Lenovo push the price up massively.

[Paid_Peanuts]

I would go hyper converged with Datacore.

 

I was quoted: DataCore SANsymphony-V Datacore HS32 - Virtual SAN with 32TB licence and one year support and it was almost £9500! Cost was then approx £1800 a year there after - seemed expensive! Even compared to VMware vSAN!

[mavhc]

Could do iSCSI Initiator for Microsoft Windows — Ceph Documentation

[deano3693]

I would buy a Dell VRTX, 3 Blade Servers, 10GB networking, local storage - mix of SSD/SATA 10k drives - We spent about 35k with a 5 year NBD warranty.

[johnpd]

I would buy a Dell VRTX, 3 Blade Servers, 10GB networking, local storage - mix of SSD/SATA 10k drives - We spent about 35k with a 5 year NBD warranty.

 

I’ve heard you can do it for nearly 19k populated (however not sure if that’s with 10gb nic). What are those dells like - nearly 4 years old now but with new blades coming out I guess to keep em fresh. VMware or hyper V?

 

Thanks

John

[deano3693]

I’ve heard you can do it for nearly 19k populated (however not sure if that’s with 10gb nic). What are those dells like - nearly 4 years old now but with new blades coming out I guess to keep em fresh. VMware or hyper V?

 

Thanks

John

 

Hi, you probably can now - Dell servers are still coping well. We brought the best processors at the time can't remember what they where and we went 128gb memory. We are about 3 years old. We've started moving some services to the cloud so are actually using less resources on it now. We're VM Ware.

 

I think i got quoted 26k for one the other week latest servers/processors and 128gb memory, 10gb nics etc with a 5 year NBD warranty.

 

Dean.

[dhicks]

If it were my money I would be frugal, grab 3 really good hosts and a veeam host second hand

 

That's a good point - I'd not factored in a separate backup server, although I guess for a refresh you'd probably have a spare older server around anyway. Also a good point above about 10GB networking for the server.

[nicholab]

Hi this is worth noting Dell’s VRTX Gains New Storage Option » Blades Made Simple. I have seen some negative commitments about HA on spiceworks about the Dell VRTX but I thought is was a great product when released.

 

I also see that RM are reselling them now as well.

Edited February 6, 2018 by nicholab

[Norphy]

When I was talking to the Dell people at BETT, they had a VTRX on the stand but they said it was legacy and will be discontinued imminently. I have no idea what it's going to be replaced with though.

[PotNoodleTech]

Problem with the VRTX is that it's a pretty massive single point of failure in one chassis. Granted with dells reliability (in my experience) you may never ever have a problem, but... if you do, your basket falls off the back of the tractor, ALL your eggs are going to be broken.

[SHimmer45]

When I was talking to the Dell people at BETT, they had a VTRX on the stand but they said it was legacy and will be discontinued imminently. I have no idea what it's going to be replaced with though.

 

really i thought it was the next big thing for dell to grow market share further in schools and smb. and would have a long life.

perhaps it hasnt worked out as well as they would have liked?

[mjk]

I'd buy cheaper and less reliable servers in school and then buy redundant servers inside AWS or Google Compute engine.

We are in a middle of this project now and get redundancy if our server room packs up or any individual server fails it will failover to Google servers.

As the price goes down of the cloud servers (it decreases year on year) we will eventually bin the local ones.

[Paid_Peanuts]

I'd buy cheaper and less reliable servers in school and then buy redundant servers inside AWS or Google Compute engine.

We are in a middle of this project now and get redundancy if our server room packs up or any individual server fails it will failover to Google servers.

As the price goes down of the cloud servers (it decreases year on year) we will eventually bin the local ones.

 

But surely, this is a problem with AWS/Azure/Google is that your interfacing with your content across a WAN link, which in most cases, can be the single point of failure and introduces other issues like latency etc.

 

I don't know what your experience of online storage is like but when we tried it here on mass with OneDrive we had complaints in all over the place that it was slow etc. We timed it to upload 200MB of content took 40 seconds. A copy of that size to our servers takes only a few seconds. It is this interaction with files that annoys our onsite users to the point where we have largely discounted online services at the moment. We have a 500MB MIA so it isn't the fact we have a saturated link - far from it. At times of testing I have done it out of hours when our connection is barely being used.

i'm still to be convinced of using AWS/Azure etc. I also found the cost to be astronomical.

Edited February 6, 2018 by Paid_Peanuts

[mavhc]

If you're scaling up and down a lot, are very large, or very small, cloud computing is good. If your users are highly distributed it's good. But I don't see why 1000 users in a building each with a gigabit link to a server with 1ms latency would find it useful.

[PotNoodleTech]

If you're scaling up and down a lot, are very large, or very small, cloud computing is good. If your users are highly distributed it's good. But I don't see why 1000 users in a building each with a gigabit link to a server with 1ms latency would find it useful.

 

Yes I see cloud storage as an additional, not a replacement too. It's taken me 15 years to get from 100mb links to multiple 10GB links - I'm darned if I am going to go back to using a 100MB [wan] link for everything!

[mjk]

But surely, this is a problem with AWS/Azure/Google is that your interfacing with your content across a WAN link, which in most cases, can be the single point of failure and introduces other issues like latency etc.

 

A redundant leased line costs us about £6000 per year, it's about the same as Microsoft EES would cost us if we used it.

Even an offsite redundant DHCP server doesn't seem to mind the latency which is about 10ms.

 

I don't know what your experience of online storage is like but when we tried it here on mass with OneDrive we had complaints in all over the place that it was slow etc. We timed it to upload 200MB of content took 40 seconds. A copy of that size to our servers takes only a few seconds. It is this interaction with files that annoys our onsite users to the point where we have largely discounted online services at the moment. We have a 500MB MIA so it isn't the fact we have a saturated link - far from it. At times of testing I have done it out of hours when our connection is barely being used.

 

We are using file stream which only streams the files as you need them, it's kind of a cached online storage and doesn't suffer from the problems that you experienced.

 

i'm still to be convinced of using AWS/Azure etc. I also found the cost to be astronomical.

 

We found GCP a lot cheaper. Also consider that you might not need to run your redundant servers 24/7. perhaps only 8hrs a day. And you don't need to run huge servers that do nothing 99% of the time. If you containerise your applications and run them from docker you will find the costs much more agreeable. You are probably trying to compare a multi-core high RAM server running windows, but building apps (DNS, DHCP, Samba) as discrete containers is the way forwards.

[mavhc]

Then you ask why do you have computers at all, and switch to serverless lambda queries.

 

But as 99% of school servers are Windows, mostly you'll just suspend them for 16 hours a day, and weekends, and holidays.

 

First step is to automate the installation of a server, once it's just a script you run, then you're in business. Infrastructure as code.

[mjk]

Can we be clear that my primary objective is failover for if the SAN, HBA or blade chassis packs up or there is a critical incident eg a fire and we move buildings. I'm talking about cloud failover and we'll only look at using this as replacement when is has proven reliability and is cost effective. That is why I put it forward as a suggestion to spending £££ on. I'm very surprised nobody else on this forum is advocating this.

 

First step is to automate the installation of a server, once it's just a script you run, then you're in business. Infrastructure as code.

This was the first thing I did ! - I've now got eight servers that are built entirely from scripts. Once this was done it is trivial to put them in the cloud, or anywhere else.

 

But as 99% of school servers are Windows, mostly you'll just suspend them for 16 hours a day, and weekends, and holidays.

 

I agree that Windows is why it is difficult for people to change but there are ways around this if you look: for example MSSQL runs well in a (linux) Docker and can presumable be setup for distributed and load balanced systems.

AWS offer a in cloud domain controller which is linux based, or you could roll your own using SAMBA4. Papercut is proven to work nicely on linux as do DNS and DHCP so these are really good choices to have a cloud based resilient backup.

 

[RusselHairy]

Speaking about replication to cloud, I would consider asynchronous replication only. Something like Storage Replica in WS2016. IMO this is the only feasible way to achieve high performance and have an ability of failing-over to the cloud in case of disaster.

 

If there is a need to deploy a highly available infrastructure on-premise, a hyper-converged approach is often the most optimal, despite the significant CapEx. For most cases, if it is possible to fit everything into two beefy servers "powered by" HPE VSA/Starwind/whatever while still having some room for future growth, I'd rather stick with it instead of buying 5+ nodes to build a capable VMware VSAN or S2D cluster. The fewer components you have, the less is a probability of shit hitting the fan.

[Roberto]

Also consider that you might not need to run your redundant servers 24/7. perhaps only 8hrs a day. And you don't need to run huge servers that do nothing 99% of the time. If you containerise your applications and run them from docker you will find the costs much more agreeable. You are probably trying to compare a multi-core high RAM server running windows, but building apps (DNS, DHCP, Samba) as discrete containers is the way forwards.

 

I think this is key. I won't claim to have fully made the transition in my own mind yet but for those of us who are used to servers on-site just being there and having resources to spare, running a server in the cloud on the same basis will prove to be massively expensive. This is a similar change of mind-set from where we stand with on-site virtual server farms, or maybe even a larger one, than what was required in order to "get" server virtualisation on site in the first place.

 

Being able to break things down so you're running more discrete 'services' on an as-needed basis is the way to go, but it does require a totally different mind-set, not least thinking hard about whether capacity in the cloud is best served by scaling up the way we typically do on-site by buying a bigger server or by scaling out by running a set of small containerised instances and automatically increasing/decreasing the amount of containers based on load; the cost of 10 small VMs can be far smaller than one large VM in both Azure and AWS.

Edited February 8, 2018 by Roberto