The_IT_Guy Posted March 19, 2015 Posted March 19, 2015 Hi I am new to the Network Manager roll and in the future we will be needing to replace some servers. But what i want to know is how do you know what spec to look at buying? e.g. how much RAM, processor, HDD type and capacity. We have three servers currently running RM CC4. Ideally i want to virtualise but can't afford it in one go so was thinking of replacing the three physical servers with three new physical servers but with a good enough spec to then the following year buy a SAN and licensing to virtualise and use them as my hosts. Will be a bit of overkill for 12 months but would allow me to phase it in. We have around 1100 users with 400 desktops and 125 mobile devices (Laptops and iPads). Any tips on how to spec this up or ideas for spec with reasons would be awesome.
Steve21 Posted March 19, 2015 Posted March 19, 2015 We're running 2 servers, 2x CPU (6 core 12 thread), 128GB RAM each. Only local HDD for HyperV as using SAN Running 36~ virtual servers on these, with enough RAM left to move most to one host should be needed if one dies etc. Steve
Steve21 Posted March 19, 2015 Posted March 19, 2015 On a side note though, unless you really want a SAN have you just considered local storage and HyperV replication etc for free? Unless you're running a lot etc wouldn't really say SAN is needed Steve 2
tmcd35 Posted March 19, 2015 Posted March 19, 2015 The question, is what are you going to use the server for? Single task, multiple tasks - if so which, virtualisation? If virtualisation how many virtual machines? What are they running? AD? DHCP? DNS? File Server? Print Server? MIS? SQL? etc, etc. For an idea of setup, we are a similar sized school (1000 users, 400 clients) and are fully virtualised. We have three high powered host servers connected to a storage array offering SMB shares (rather than a SAN). To give an idea, we tend to go for either Novatech or VeryPC custom builds and spend no more than £2500 on a server. Because they are bought for virtualisation we get as many cores and as much ram as we can squeeze out of supplier (last server was 24 core and 128Gb, I think) and just enough hdd space for the OS. Our storage array cost us £7,500 and holds around 8Tb of SAS storage. If you are looking to buy three servers for virtualisation I'd say budget around the £7,500 mark. Forget the SAN and use local storage with Hyper-V replica running between the three hosts for failover/redundancy. For the £2500 per server mark I'd be looking at 8-12 cores, 48-64Gb Ram, and 1.5-2Tb storage min in each, possibly using tired storage and mixing SSD, SAS, and SATA and getting 4-6Tb in each server (1-2Tb SATA drives are cheap). 2
DavePa Posted March 19, 2015 Posted March 19, 2015 Another factor that is sometimes overlooked is the expected/desired lifetime of the server - this usually includes looking at the immediate HDD requirement and factoring in a growth factor. I've seen servers purchased with 2tb HDD space to replace old servers that also had 2tb with 1.9tb in use...not going to last the expected 5 years without an upgrade of some description!
The_IT_Guy Posted March 19, 2015 Author Posted March 19, 2015 As i am a bit new to this i am being a bit guided by suppliers (a bit dangerous i know). Didn't know you could virtualise without a SAN. So with Hyper-V replica it still allows you to take advantage of redundancy? How does that work? Obviously with a SAN the files are stored on there and the hosts all access them. How does it work with just local storage? In terms of if there is a failure to one of the hosts? In terms of what they will run it is our curriculum network including DHCP, DNS, user directories, Print Server, WSUS, WDS, AD. Not entirely sure on the final total of VM's as it would need to cover these services and then gradually adding our VLE (Moodle) and our MIS server to it at a later date. I basically have about £20k to move from CC4 to a vanilla network but been given the job last minute and just seeking advice for the best way to do it to get value for money and a strong network.
Steve21 Posted March 19, 2015 Posted March 19, 2015 Didn't know you could virtualise without a SAN. So with Hyper-V replica it still allows you to take advantage of redundancy? How does that work? Obviously with a SAN the files are stored on there and the hosts all access them. How does it work with just local storage? In terms of if there is a failure to one of the hosts? Effectively the hosts copy the files in the background between eachother, so one has a "live" machine, and one has "spare". If one dies can flick other on etc. Steve 1
tmcd35 Posted March 19, 2015 Posted March 19, 2015 I basically have about £20k to move from CC4 to a vanilla network but been given the job last minute and just seeking advice for the best way to do it to get value for money and a strong network. You can do a lot with £20k! 3 decent servers and a SAN easy. Best people to talk to are @VeryPC and Novatech, IMHO. They can work wonders on that sort of money. The most expensive part of all is the physical hard drives, but with creative tiering that doesn't necessarily need to be as expansive as you first think (you don't really need fill a SAN with 15k SAS drives - a mixture of SSD and SATA with fewer SAS might give better performance and more space for less money). The starting point is your existing setup. What are the specs and roles of the existing servers? Do you want to virtualise 1:1 or do you want to split roles on to their own servers? (eg. WSUS and WDS on the same server, or 1 WSUS server and 1 WDS server). 3
ICTDirect_Dave Posted March 19, 2015 Posted March 19, 2015 £20 Grand? Crikey thats quite a lot! I was just going to suggest some of my lovely refurbished servers might well do the trick, but I'd struggle to take even half of your money when speccing up a few superbly specced servers
featured_spectre Posted March 19, 2015 Posted March 19, 2015 work on the principle of this as a minimum (its how I used to do it) 2 cores per VM 16GB RAM per VM 2TB HDD per VM Then you will need to think about SANs and/or NAS's depending on need.
tmcd35 Posted March 19, 2015 Posted March 19, 2015 work on the principle of this as a minimum (its how I used to do it) 2 cores per VM 16GB RAM per VM 2TB HDD per VM How much!?! Most VM's barely need 1 core, SQL servers being the exception where 4 core is probably fine. 4Gb RAM is probably over generous for most VMs, but hey if your running Win2012r2 with full GUI in each VM you can chalk that up to Windows bloat. Again only SQL needs anywhere near 16Gb. As for HDD space. 200Gb per VM is probably more realistic. Refer to your DB provider for their requirements (our SIMS server only has 250Gb). Only your file server will need TB of space. We're running around the 4TB for file services and 4TB for all other VMs on our storage array. We have about 30 VMs and the storage array is a little under half full. 2
featured_spectre Posted March 19, 2015 Posted March 19, 2015 Only reason I spec so high is because of redundancy and most were acting as file servers for individual year groups (at the request of the Director of ICT at the time) so its what I did.
Steve21 Posted March 19, 2015 Posted March 19, 2015 All our VMs only run 1 core (2 for file server or higher e.g. SQL/SIMs), and 4GB RAM dynamic (16 for SCCM/SIMs), 150-200GB per machine unless file server etc. Guess it depends what you have on the servers though We run lots of lil ones, as prefer the options it gives you etc. Steve 1
tmcd35 Posted March 19, 2015 Posted March 19, 2015 When spec'ing host servers the general rule of thumb is to have them run at (n-1)/n capacity were n is the total number of hosts on the network. So, ideally, with three hosts each server should be running at 2/3rds capacity under normal loads. This then factors in redundancy for failover should one of the hosts go off line. So you need to work out how many VMs total you are likely to run, and what the spec of each VM needs to be based on its anticipated work load. Then split the VMs equally between hosts. This should then give you an indication on the required spec's of the host servers. In terms of future proving, I'd probably look to run each host at closer to 1/2 capacity under normal loads (gain assuming 3 hosts) to give room for adding new VM's later as the need arises. 1
VeryPC_Ed Posted March 19, 2015 Posted March 19, 2015 (edited) If you are looking at P2V / Consolidating existing workloads, I would recommend first of all performing an analysis of what you have currently and then work from there to determine what you need. An quite commonly overlooked tool is the Microsoft Assessment and Planning Toolkit - this will allow you to inventory your estate, then perform a performance analysis to see what CPU/RAM/Network and importantly - what IOPS your current infrastructure is using. Then from this, you can get some more insight to then look at selecting a specification and subsequent solution. For the actual virtualization side of things, there are a whole host of options available, from simple replication using Hyper-V Replica / Veeam Replication jobs to Windows Failover Clustering using a NAS/SAN or one option we can provide is a 2 Node HA Cluster (Compute and Storage) using Starwind Virtual SAN for Hyper-V. I'd be more than happy to have a chat with you to try and work out what you need and then look at some options. Thanks, Ed Edited March 19, 2015 by VeryPC_Ed 4
Stuclark Posted March 19, 2015 Posted March 19, 2015 We have 500 virtualised desktops; and they are hosted on 5 servers with a total of 320 cores and 1920GB RAM (64 cores & 384GB RAM in each server). The servers all boot off an SD card and everything lives on our SAN. (sorry, but SANs are the only really sensible solution for reduncancy). We have another 4 identical servers to host our virtualised servers. What this should tell you is that the spec you need is the highest you can possibly get - never, ever, ever work down to a price; instead look at what you need and then work out how to make the finances work (like buying in pieces, as you've said)
maniac Posted March 19, 2015 Posted March 19, 2015 A well constructed "business case" for the infrastructure change and you could find you'll have all the money in one go. One thing I have learnt in 12 years of working in schools is they *always* manage to find money if they perceive they need it. 1
tmcd35 Posted March 19, 2015 Posted March 19, 2015 We have 500 virtualised desktops; and they are hosted on 5 servers with a total of 320 cores and 1920GB RAM (64 cores & 384GB RAM in each server). The servers all boot off an SD card and everything lives on our SAN. (sorry, but SANs are the only really sensible solution for reduncancy). We have another 4 identical servers to host our virtualised servers. I think the response here is - most of us aren't blessed with that kind of money, and thankfully the OP is talking about server virtualisation not deskop. £20k should, in the hands of a reputable supplier, get the OP everything they need and be left with some change. Also, SANs are certainly not the only sensible solution for redundancy, but they certainly are a sensible solution.
spc-rocket Posted March 19, 2015 Posted March 19, 2015 Hi, I think the other people have covered all the points very well so i won't go into that. One thing that you may want to do is NOT to reveal how much you got for the project to suppliers. Just get quotes and discuss the options with them etc, if you work to a figure then it may not be best. If speaking to suppliers don't state that i have xxx amount for this project. Golden rule that hasn't let me down. Ash.
fathead Posted March 19, 2015 Posted March 19, 2015 If you're fully virtualising, either with two hosts + DAS (replicating) or two hosts + SAN (clustered), I'd also have a physical read only domain controller outside of your setup for a belts and braces approach. Or am I just being old fashioned?!
plexer Posted March 19, 2015 Posted March 19, 2015 Only reason I spec so high is because of redundancy and most were acting as file servers for individual year groups (at the request of the Director of ICT at the time) so its what I did. Except it does nothing for redundancy.
The_IT_Guy Posted March 19, 2015 Author Posted March 19, 2015 £20 Grand? Crikey thats quite a lot! I was just going to suggest some of my lovely refurbished servers might well do the trick, but I'd struggle to take even half of your money when speccing up a few superbly specced servers I should mention as well as the hardware we are wanting to try get some support / consultancy with setting up the network. Ideally someone to help install and set up the hardware and get us up and running with everything we need really so that we are not starting from nothing. Basically help us migrated onto a working network with an image for our machines etc. Just don't have the time or team to get it running ourselves.
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now