How complex is your network?

[mhosker]

Hello, First post.

 

The last school network I had involvement in a few years ago was supposed to have redundancy between two servers using DFS, this never ever worked, but got me thinking...

 

The school was private, with about 400 students, 3 buildings, so very small.

 

What is a typical level of redundancy and complexity for a school network of this size? Or even larger...

 

If it were me I would be wanting to use a couple of SANs each in different buildings using a Hyper-V failover cluster and block level replication on the SANs.

 

But is that too overkill for the scenario I describe? Using SANs is adding another level of complexity, especially if they are FC.

 

In this particular case the network was totally flat with no VLANs, so I would also want to change that and maybe split between buildings.

 

It does also get me thinking, are any of you running OSPF in your network? Or even have your own address space and are BGP peering?

 

Mike

[FN-GM]

For that size of network I wouldn't use OSPF. It simply doesn't need it. What are you trying to achieve? I would use VLANS though.

 

I would look at Storage Spaces Direct instead of a SAN. You could have 2 HyperV servers in seperate buildings, replicate the data across them and have full HA.

[mhosker]

Hello,

 

Not actually trying to achieve anything as I no longer have any involvement in the network. Just a theory question.

 

I will have a look at storage spaces, heard of it before but never had reason to use it.

 

At what point would you use OSPF? Can you maybe even give any examples of a successful OSPF deployment you have done?

 

Many thanks,

Mike

[FN-GM]

Hello,

 

Not actually trying to achieve anything as I no longer have any involvement in the network. Just a theory question.

 

I will have a look at storage spaces, heard of it before but never had reason to use it.

 

At what point would you use OSPF? Can you maybe even give any examples of a successful OSPF deployment you have done?

 

Many thanks,

Mike

Personally if it is possible I would pick EIGRP over OSPF. When you have multiple OSPF areas it can be a real pain.

 

Anyway I would use something like OSPF or EIGRP when you have lots of layer 3 devices such as switches and routers that connect via a redundant route. This might happen if you connect lots of physical sites that have their own layer 3 switches, although if you have no redundancy I would use static routes. It might also be used if your client switches are Layer 3. You can then host the VLAN on that switch and not the core. If you interconnect the client switches traffic could be routed without going through the core. Infact in some deployments you might not even need a core.

[Rob_D]

We're on the bigger side (1500) pupils 3 (and a bit) buildings.

No SANs, but we have 3 virtual hosts with the servers syncing between them (each host has some active and some backup servers).

We use VLANS but vlan by system rather than buildings. So one for phones, one for CCTV, one for BMS, that kind of thing. The networking guys I've spoken to about it say VLANing by building say it's not worth It unless you're really big or running multiple campuses.

[Jcx500]

I keep it simple one vm host

[DrCheese]

We currently have 3 VMware hosts connected to a SAN.

 

I'm moving away from the Host/SAN model tho, as it's a single point of failure (Whist heavily redundant itself, it's still a risk) & costs quite a bit more than 2/3 hosts with VM's balanced/replica's between them.

[TwistedHelixis]

All the schools I look after have about 300 / 600 pupils and I removed all the redundancy (extra servers etc) in favour of a good backup / restore. Being able to restore a vm in a few hours is good enough for a school of that size and removes a lot of complex issues that might arise. In fact since doing this I have reduced my server issues by 99%.

 

EDIT - I now have one Host with 2 VMs and an iscsi backup server that backs up the VMs.

Edited June 11, 2019 by TwistedHelixis

[DrCheese]

Yes, my logic too. We've previously had the mantra of high availability, when in reality what we actually need is quick restore. If I can be up and running from a replica within a few minutes then all the complexity needed for HA to work properly is removed.

[dmj]

My thoughts are that the best way for redundancy AND disaster recovery is to segregate the server data from the server configuration.

Every production service that gets deployed has the configuration written as code which is then stored in version control. The data is backed up and stored separately.

This has a few advantages: because the config doesn't need to be applied manually, the servers can rebuild themselves. As the servers as ephemeral they are pretty much designed for load balancing from the outset. It's really easy to see who's made a server change by looking at the git history!

[thebatman]

We have 4 big buildings and a few small building, we have split the building over two VLANs for "normal" traffic. Then we have VLANs for services IE CCTV, VoIP, Servers and switch management.

The buildings are split over two VLANs because we have two "Core" switches. This helps are the two areas are run from two substations that we often have issues with.

No redundancy as such, but the way the network is setup, we can have one or both of the building back up and running in a few hours.

[PotNoodleTech]

I always adhere to the KISS principle our old DT teacher used to swear by - "Keep It Simple, Stupid".

 

What he meant was, the more complicated you make something, the higher the chance of one of the over complicated parts going wrong.

 

So for me, there are not many secondary's big enough to warrant much more than:

 

2 x High Spec VM hosts, Hyper V or VMware. Chock full of RAM and SSDs. Use local storage or storage spaces direct.

1 x Veeam Host chocked full of 4tb - 8tb SATA drives.

1 x Secondary backup target (NAS Box) with a few 4-8tb HDDs.

Vlans are a must, I've seen schools with vlan overload, one for every IT suite, but that's.... overkill. Typically you'd want 5-10 vlans (Management, Curric, Admin, CCTV, Phones, Wireless).

 

What I am trying to say is don't over complicate the matter, all you end up doing is making things harder for yourself and making things cost more for the school. You can achieve a vry fast network with 99.9+ uptime with the above simple system at a cost of ~£20k every 5 years (less of you use refurbished equipment, though I do always buy new server SSDs).

 

My 2p .. many others will disagree!

[Oaktech]

I'm a halfway house between simple and complex. I have a minimal number of VMs - one file server VM with a robust backup, one V-DC, one physical DC, one print server, one RDS gateway, 2 RDS hosts, one Direct Access, One Utility (with WSUS & SCCM & Power Management) as simple an AD structure as I can get away with, one Physical backup server. Only 3 VLans (default, management and phones) but it's all built on a San with a pair of Hyper V clustered hosts.

[TechMonkey]

I always adhere to the KISS principle our old DT teacher used to swear by - "Keep It Simple, Stupid".

 

What he meant was, the more complicated you make something, the higher the chance of one of the over complicated parts going wrong.

 

So for me, there are not many secondary's big enough to warrant much more than:

 

2 x High Spec VM hosts, Hyper V or VMware. Chock full of RAM and SSDs. Use local storage or storage spaces direct.

1 x Veeam Host chocked full of 4tb - 8tb SATA drives.

1 x Secondary backup target (NAS Box) with a few 4-8tb HDDs.

Vlans are a must, I've seen schools with vlan overload, one for every IT suite, but that's.... overkill. Typically you'd want 5-10 vlans (Management, Curric, Admin, CCTV, Phones, Wireless).

 

What I am trying to say is don't over complicate the matter, all you end up doing is making things harder for yourself and making things cost more for the school. You can achieve a vry fast network with 99.9+ uptime with the above simple system at a cost of ~£20k every 5 years (less of you use refurbished equipment, though I do always buy new server SSDs).

 

My 2p .. many others will disagree!

 

Don't disagree with you in principle but would with blanket statement and number of hosts. Both places I have worked with Virtual infrastructure have had 3 hosts as not everything would run on one and three gave you redundancy to have one go down or taken down for maintenance with no, or minimal, issues. N+1 principle.

 

Beyond that it is all personal preference. We have a virtual Veeam host that backups to a NAS and a secondary NAS. I prefer a SAN to reduce issues (although it adds complication I think it is worthwhile) but I'm going to be looking at Hyperconverged next time round.

[DrCheese]

I looked at hyperconverged but it either came out too expensive or overly complex for what we were trying to achieve - Only issue with the SAN model is that it's an eggs in one basket setup, i.e if the SAN dies for whatever reason you're screwed unless you have a replica on another device of your core services.

[Jcx500]

All Local Storage Didn't see the point of a San, Raid 1 for OS drive and raid 6 for hyper-v

[TechMonkey]

I looked at hyperconverged but it either came out too expensive or overly complex for what we were trying to achieve - Only issue with the SAN model is that it's an eggs in one basket setup, i.e if the SAN dies for whatever reason you're screwed unless you have a replica on another device of your core services.

 

Yes, this is a consideration but with dual controllers, dual power supplies and cross connections it is highly unlikely. I can never see the difference between that argument and using local storage on a host. My understanding is it is a harder migration if a host needs to be taken down as you have to migrate the VM and then the data. Also using local storage can you still bring VMs up if a host fails? That may be a misconception of mine though.

[DrCheese]

Yup - I think you just need to have a plan in place to bring services backup if the SAN dies completely - Most places will have backups but not necessarily anywhere to restore the huge VMs to. A older server crammed full of disks will probably be enough to bring up the basics whist the SAN is AWOL

[LTurner3692]

This is where I personally think Azure can come in handy. We do a back-up on site, and a VM back up offsite using Veeam, If we were to lose a building or have a major issue, id simply bring it back up in Azure. We also use Cisco Meraki, this connects all our sites together without the need for a P2P link.

[Oaktech]

Yup - I think you just need to have a plan in place to bring services backup if the SAN dies completely - Most places will have backups but not necessarily anywhere to restore the huge VMs to. A older server crammed full of disks will probably be enough to bring up the basics whist the SAN is AWOL

 

Yup, my old servers have been given sufficient space and horsepower to operate a reduce functionality, reduced resiliency, network while the SAN is repaired if necessary. They got an upgrade from 1TB sas (in 4x 250gb) to 8TB sata (in 4x 2tb) storage and memory hike from 16 to 32gb ram and then got repurposed into non critical roles. I can pull them out and restore to them if necessary.

[titch]

This is where I personally think Azure can come in handy. We do a back-up on site, and a VM back up offsite using Veeam, If we were to lose a building or have a major issue, id simply bring it back up in Azure. We also use Cisco Meraki, this connects all our sites together without the need for a P2P link.

Out of interest how easy was it to set up so that you can spin up in azure? Are there many real world prerequisites? What would you need to do internally to get everything to connect via azure should you lose your onsite gear.

[LTurner3692]

Out of interest how easy was it to set up so that you can spin up in azure? Are there many real world prerequisites? What would you need to do internally to get everything to connect via azure should you lose your onsite gear.

 

Unfortuntaly it was put in before i started, ill ask the company how easy it is to do(i have a meeting with them in the next week). However its one of them things that now i have it i wouldn't let it go for all the money in the world.

[CCCSecMan]

All Local Storage Didn't see the point of a San, Raid 1 for OS drive and raid 6 for hyper-v

 

Agree. This would be how I would do it now for a secondary school size environment. 2 or 3 hypervisors with RAID1 SSD storage with capacity for 1 failure. No SAN. Daily backups with a documented and quick recovery procedure - worst case if a server fails and the storage is toast too just restore all VM files to remaining hypervisors and load from there. Save thousands by not requiring a SAN and could spend that on getting high spec SSDs with ample capacity. Growth simple too as you just buy an extra server with storage as and when.

[howartp]

I looked at hyperconverged but it either came out too expensive or overly complex for what we were trying to achieve

We just tendered for 3 hosts, FC SAN, SAN switches and replacement core switches.

 

Quotes were £35,000 to £90,000 all in.

 

The £35,000 was a Dell VRTX that met all the specs of 3 hosts, SAN and Switching, plus separate core switches.

[PotNoodleTech]

The £35,000 was a Dell VRTX that met all the specs of 3 hosts, SAN and Switching, plus separate core switches.

 

One chassis to rule them all, once chassis to find them, once chassis to fail catastrophically, and in the downtime bind them.