Virtualization project

[stevewill6]

Good Morning everyone.

 

With server 2003 support up soon and some small issues with our three 10 year old servers it’s time to seriously look at upgrading to a virtualized setup.

 

A little background info - We are a secondary school with 700 users and 350 devices. 1 LEA server (User data, DHCP) and 3 of our own (WSUS, Deployment, Print and some other bits and pieces)

 

I am planning on replacing our 3 servers with two hosts (Hyper-V) both with local storage. Also there is a old server available for a Veeam backup server.

 

Initial spec

2x Intel Xeon Quad/Hex

64 - 94 GB RAM

8* 300gb SAS 10K (Raid10) 1.2TB usable storage

 

Questions

 

I want some type of redundancy on the hosts so if one fails the other will pick up the slack. Best way to do this? Hyper-v replica or 2 host Cluster?

 

Recommendations on what switch to connect everything together?

 

Would 4 Network ports be enough - 1 management, 1 replication/cluster, 2 for vm's.

 

Thanks, Im sure ill come up with a few more questions.

 

Edited November 28, 2014 by stevewill6

[Net-Ctrl]

Hi @stevewill6 i'm afraid i don't have any answers for the above, and to make matter worse, i wanted to ask a question. What are your plans to secure your virtualised environment?

[Steve21]

First thing that springs to mind is I'd say more RAM the better In reality if you're using it as a cluster and want full failover ability you're limiting yourself to 32GB per box (at 64GB)

 

In terms of local storage not sure how you'd work that into redundancy (With cluster) as if you're storing a VM on one HOST and it goes down, there's nothing to failover to as you don't have the VM files to access (Obviously with SAN etc it's on the SAN that both are linked to)

 

Steve

[zag]

Good Morning everyone.

I want some type of redundancy on the hosts so if one fails the other will pick up the slack. Best way to do this? Hyper-v replica or 2 host Cluster?

 

 

First question is why do you need instant redundancy? With virtualization and veeam backups you could easily restore a host in 20mins.

 

I can understand if you are operating a datacenter needing instant fallover, but in a school its way overkill imo.

 

Also sounds to me like you are putting all your eggs in one basket with only 2 hosts.

 

We bought cheaper servers and have about 7 physical here, running 18 hosts. Works well and easy to restore any VM to another host.

[dhicks]

In terms of local storage not sure how you'd work that into redundancy (With cluster) as if you're storing a VM on one HOST and it goes down, there's nothing to failover to as you don't have the VM files to access (Obviously with SAN etc it's on the SAN that both are linked to)

 

I thought that was the whole idea of Windows Storage Spaces - it handles replication between nodes, so anything available from one node is also available on the other? The setup as described by the original poster would seem to be just right for that kind of setup, although you'd get the same issue with disk storage space as you've pointed out with the RAM - you'd be replicating each VM image between machines, so each server's 1.2TB of storage space would have to hold a full set of VM images for both machines. We have a similar setup ourselves, although using Xenserver with DRBD for replication, and 1.5TB(ish) on each server. It's just about enough at the moment, although a bit more storage space is always handy for future-proofing. Might SSDs be a realistic option?

[Norphy]

First question is why do you need instant redundancy? With virtualization and veeam backups you could easily restore a host in 20mins.

 

I can think of two big, big reasons why failover is desirable.

 

1) Your scenario assumes that either your backups are up to the minute or that the data loss between the last backup and the time of the system failure is acceptable. It may (and only may) be acceptable for user data, losing a days work wouldn't be the end of the world but if there's any financial data being hosted there, you may end up in real trouble.

 

2) It also assumes that downtime is acceptable. 20 minutes might not seem like long but if it's your cashless catering system that's gone belly up during lunchtime, it'll seem like an eternity. If it's your exams systems that's gone down, students aren't going to be happy to hear "Oh, give me 20 minutes to get the system restored and then give invigilators another half hour to get the exam set up again".

 

Failover isn't overkill, it's sensible.

 

I do agree that two hosts may not be enough though.

[dhicks]

First question is why do you need instant redundancy? With virtualization and veeam backups you could easily restore a host in 20mins.

 

That 20 minutes is assuming you're sat at your desk, ready to go, and that there's not a constant stream of people asking you when stuff will be fixed. In practice, I'd go for instant failover for if something fails overnight, or if you just want a less stressful time when stuff conks out. I'd aim to consolidate into two physical servers if possible, too, just like the original poster seems to be planning. Microsoft's Hyper-V with VM images stored on a Windows Storage Spaces volume does seem to be pretty much aimed at eactly this kind of setup - smaller scale than a huge great datacentre setup, but still offering good failover capability. Our Xenserver / DRDB setup does work but is rather clunky in operation - rebooting a server is a rather variable process as something in the DRDB setup sometimes causes connection issues, if I was starting from scratch I think I'd probably go for pretty much the hardware setup the original poster describes with Hyper-V running on a Storage Spaces volume. It'd be best to keep those two servers physically separated if possible, or even get one off site into a datacentre on in another school, but that depends on budget / connectivity, of course.

[Steve21]

I thought that was the whole idea of Windows Storage Spaces - it handles replication between nodes, so anything available from one node is also available on the other?

 

Honestly not looked at them that way. I thought the point was using a separate tray you effectively turn that into the SAN via Storage Spaces, if you can do it that way I guess that works then

 

Guessing you might want more disks though then if you're stealing some for OS, and some for Storage Spaces.

 

Steve

[SHimmer45]

Hyper-V replica works well but isn't an auto switch over if something breaks, but your machine states will be on another server and just need to be turned on (which for some people who dont have a cluster/san gives you redundancy)

for clustering the recommended min of network ports is 3 i think

 

Storage spaces does open up the ability to turn a server into a SAN type device without the massive outlay which has to be a plus

[dhicks]

Guessing you might want more disks though then if you're stealing some for OS, and some for Storage Spaces.

 

That's a good point. I don't know how much disk space a minimal Hyper-V install takes up, but XenServer only uses 4GB partitions, I'd have thought a 32GB SSD for the OS disk in each machine would be fine, then use all the main storage space in each server for VM image storage. I should maybe add that our 1.5TB for VM images doesn't include user files - those are on a separate 12TB array.

[Steve21]

That's a good point. I don't know how much disk space a minimal Hyper-V install takes up, but XenServer only uses 4GB partitions, I'd have thought a 32GB SSD for the OS disk in each machine would be fine, then use all the main storage space in each server for VM image storage. I should maybe add that our 1.5TB for VM images doesn't include user files - those are on a separate 12TB array.

 

FYI 45GB currently on our hosts, as it's Server 2012R2 all patched up etc But yes, either way needs to be considered, and then it's whether you want those RAID etc.

 

Steve

[PotNoodleTech]

Hi @stevewill6 i'm afraid i don't have any answers for the above, and to make matter worse, i wanted to ask a question. What are your plans to secure your virtualised environment?

 

SELL SELL SELL baby!!!

[PotNoodleTech]

I would suggest the possibility of 3 lower spec hosts rather than 2 high spec hosts.

 

Can't comment on the redundancy aspect as I'm not a HyperV man but avoiding a SAN is a good idea. VMware lets you do this with VSAN - but not sure if HyperV has the same kind of functionality.

[Net-Ctrl]

SELL SELL SELL baby!!!

 

It was meant as a constructive question, promise

 

I'm interested to hear how schools are securing their virtual environments. We do have a solution for it, but it's always good to know the other routes schools are taking with it.

[stevewill6]

Hi Thanks for your reply's.

 

I haven't been clear with regards to redundancy. Its not instant that is required sorry. We would be quite happy with 20 minutes even an hour of down time to recover. I suppose what I was asking is what would be the best way of achieving that?

 

I will look in to running more cheaper servers but I can see licencing costs increasing with that? we don't have much of a budget for this project.

 

I will have a look at storage spaces if that is capable of giving us that instant fail over as it would be great.

 

Disk space is a concern but I have estimated all the vms I would like to run should be about the 800gb mark across the 2 hosts. This leaves us room for a few more in the future but I was looking to add hosts if we needed more capacity.

 

In what way do you mean secure them? They will be locked up in the server room.

[tmcd35]

Good Morning everyone.

 

A little background info - We are a secondary school with 700 users and 350 devices. 1 LEA server (User data, DHCP) and 3 of our own (WSUS, Deployment, Print and some other bits and pieces)

 

 

We're a very (very) similar sized school and are fully virtualised. As ever, if you are anywhere near Norfolk then you are always welcome to drop by and chew the cud.

 

 

Initial spec

2x Intel Xeon Quad/Hex

64 - 94 GB RAM

8* 300gb SAS 10K (Raid10) 1.2TB usable storage

 

My thoughts are - processor cores, ram, and disk space. How many VM's do you plan to run? What tasks do you want each to do? What are their individual requirements. Our latest host server has something like 24-cores and 128Gb Ram. 1.2Tb seems very low. We are struggling with 4.5Tb and have just put in plans to double that over the next couple of months.

 

I want some type of redundancy on the hosts so if one fails the other will pick up the slack. Best way to do this? Hyper-v replica or 2 host Cluster?

 

That's a cost/benefit question only you can answer. We're going down the Replica route because its (a) cheap and (b) we don't need instant failover. Also, I don't think you can cluster without a SAN set up, atleast not for auto failover. The VM's need to be on shared storage that both hosts can see and access. Replica gets around that problem at the expense of having 100% live migration and instant failover.

 

Would 4 Network ports be enough - 1 management, 1 replication/cluster, 2 for vm's.

 

We have 1 management, 2 to the shared storage, 2 for the VM's virtual switch.

Edited November 28, 2014 by tmcd35

[stevewill6]

Rough list of the Vm's I will be running.

 

Print management server

Dual Core – 4gb RAM – 100gb Drive – Win Server 2012r2

Deployment Server (Ghost / SCCM / WDS-MDT)

Dual Core – 6gb RAM – 300gb Drive – Win Server 2012r2

Spice works / Classroom Management application Server

Dual Core – 4gb RAM – 100gb Drive – Win Server 2012r2

Webserver (Intranet, Wiki)

Dual Core – 4gb RAM – 50gb Drive – Linux

 

Plans to add an extra two in the short term

 

Network Monitoring Server

Dual Core – 4gb RAM – 100gb Drive – Win Server 2012r2

Library application server

Dual Core – 4gb RAM – 100gb Drive – Win Server 2012r2

[tmcd35]

I'd assume from that then - your DC's, file servers, MIS, etc are all seperate and not being virtualised at this time? Would there be plans to virtualise them in the future?

 

Based on that list of servers I'd say, use a hex cores with HT, 64Gb should be plenty of ram, for each server. Hyper-V replica will do what you need.

 

The only server I'd question is the deployment server. SCCM is a pig. I think ours is currently set to 4 cores and 16Gb of Ram, using 14Gb idle!

[stevewill6]

Yes, DC, and SIMS are off site with the LEA. User date in stored on a LEA server onsite. No plans to add these to the project for the next few years.

 

We currently use Ghost to deploy images but i'm hoping to move away from this to SCCM eventually.

[Norphy]

There will be those who disagree with me entirely here but I'm going to say it anyway. I don't think that SCCM is a particularly good candidate for virtualisation.

 

First of all, it's huge. It needs a SQL database. Storing OS images and software deployments takes up gobs of space. If it's sharing resources with other servers, the performance of those servers is going to suffer when it's pushing out OS images or updates or whatever. If I were you, I'd put it on its own server if you can.

[tmcd35]

^I agree with this entirely. The only reason it's virtualised here is because the spec's of our host servers can easily support it with more than enough room to spare.

[smithson83]

I have a similar setup to how you described in the first post, but I took the route to spec a server that would be able to run EVERYTHING and then some, then get two, have half VM's running on one Host and half on the other, replica between the two, then backup up one host offsite in case the building burns down.

 

Each host is Currenly 2x Xeon Quad 2.0Ghz, 128GB Ram, 10x300GB SAS 15k, 2x(4x1GB)NICs, Win Svr 2012R2 DC

Backup: 6TB Nas across site, 5 2TB USB-HDD(one Taken offsite each night)

 

The 800GB file server can do a full live migration in 30Mins (or restore from the NAS), each server is replicated ready to go live if other host fails, you could script to check the heart beat of the live VM and start the replica if its down for xx mins, but I prefer to handle it personally, only a few clicks

[Ephelyon]

Give @ICTDirect_Dave a call. When I was last looking at their HP DL585 G6 pricing, we were able to get something like 24 cores, 128GB RAM and 3TB storage in RAID10 for about £1,325/unit.

[ICTDirect_Dave]

Give @ICTDirect_Dave a call. When I was last looking at their HP DL585 G6 pricing, we were able to get something like 24 cores, 128GB RAM and 3TB storage in RAID10 for about £1,325/unit.

 

And then there's the wonderful little 1U HP Special edition machines we took from a microsoft data centre. Pair of hex cores (X5670) 96GB 10600R & 8 x 300GB 10K SAS for less than two grand. Awesome machines!

[PotNoodleTech]

And then there's the wonderful little 1U HP Special edition machines we took from a microsoft data centre. Pair of hex cores (X5670) 96GB 10600R & 8 x 300GB 10K SAS for less than two grand. Awesome machines!

 

Sounds noisy!