If you had to spec out a brand new server room

[Trapper]

What would you get?

 

I'm trying to get with three decent spec'd HP servers to run Hyper-V instances. Two in use, one as a hot migrate to. All connected to a storage array where the VMs live.

 

Would you do this? If you would what would you choose? Assume 4/5 VMs per host.

 

Not that I have to make this a reality in .... 3 weeks!

[twin--turbo]

4-5 VM's per host, that is a very low count.

 

Buy two machines with extra memory/cpu so that you can run all guests from one server in the event of failure.

 

TT

[Trapper]

That's a plan.

 

Luckily we have enough money to mirror the set up at the other two sites. Yes the annex sites will run at lot less, but if I can get the replication right it might be a pretty good way to sort out backup!

[Steve21]

If they're all connected to the live storage, why not use all 3 as active? Can still replicate between etc, and if you're using shared storage I'd assume you're doing cluster so no need for a "spare" as such

 

As above 4/5 seems very low, we run 20+ per HOST and still spare room if we don't take replication into account

 

Steve

[Trapper]

What would you use for storage? It's been a very long time since I last bought enterprise class storage! So long it was a SCSI connection!

[Steve21]

What would you use for storage? It's been a very long time since I last bought enterprise class storage! So long it was a SCSI connection!

 

Depends what you're thinking of doing. If it's cluster based it's going to be some kind of iscsi setup (Unless you're faffing with storage spaces etc).

 

But if you're just wanting replication and have a live DR room anyway you could always just use local storage on the servers, makes life a lot easier and less hassle

 

Steve

[NB9457]

What would you use for storage? It's been a very long time since I last bought enterprise class storage! So long it was a SCSI connection!

 

I'd use SSDs, price has plummeted so much over the last year that I'm confused when we are still installing Servers and PCs in schools with mechanical drives!

[Trapper]

So you wouldn't go for a SAS device? I assume it's overkill...

 

DAS is a bit useless with two servers, unless one serves it via iSCSI but that's a bit faffy

 

Problem is we could afford two high specced servers with fibre channel and a SAS with 2.4TB well under budget, but I feel it's just wasteful.

 

We might run 10 VMs total - max of 250 kids across three sites (special) from 3-19. We've never done anything anywhere near like main stream secondaries do.

[Steve21]

So you wouldn't go for a SAS device? I assume it's overkill...

 

DAS is a bit useless with two servers, unless one serves it via iSCSI but that's a bit faffy

 

Problem is we could afford two high specced servers with fibre channel and a SAS with 2.4TB well under budget, but I feel it's just wasteful.

 

We might run 10 VMs total - max of 250 kids across three sites (special) from 3-19. We've never done anything anywhere near like main stream secondaries do.

 

I'd stick with SAS personally, SSD is great if you need the IOPS but I'm assuming you don't by the sounds of it. Guess how long do you want to future proof it for?

 

Also in regards to DAS you'd just have both servers with storage, then replicate across, no need for shared storage unless you're doing cluster? Which then would make the third-hot one pointless as you'd either have all 3 live, or wouldn't need the 3rd etc

 

Steve

[Trapper]

I've been struggling with this for months now. They've given me a huge budget, just under £20k for the new secondary site. Which will house 100 kids max as KS3/4 age. As we're a special we really do not use IT in the same way a mainstream does. No video editing, no media work, mostly Office, internet and some locally installed apps.

 

I've taken to thinking of our three sites primary, secondary and Post-16 as in effect using IT like three one form entry primary schools. My SLT expects the new secondary site to be all singing and all dancing. Massive iron powering mega-jigga-terra flops. When really all we need is 5/6 VMs to run AD, apps 1, apps2, MIS etc and a few TB of storage space.

 

The primary and P-16 sites require less. 3/4 VMs max at each site with the same storage for DFS to replicate everyone's data.

 

It's been difficult to say "No, I really do not need a server room our local 1500 pupil secondary school would beg for, all I need is £5/6k to sort everything out". I also hate to think of the Head's expression when he looks into our new server room and see's one box!

[PotNoodleTech]

To be honest, if the headteacher wants you to spend decent money, and ensure a 100% reliable and fast server infrastructure rather than skimping and compromising to meet a tight budget like most of us unfortunately have to then he is wiser than many! Please take him up on the offer. Don't waste money though, just spend it in the areas which will make maximum difference.

 

Personally, I would go for two very well spec'd servers with 5yr warranty plenty of SSD local storage, preferably spare disk slots for future expansion, run hyper-v on it and configure replication so if one goes down the other can take over. This will give you 99.9% uptime probably and it saves you the hassle and risk of shared storage (the shared storage, san/nas/jbod tray becomes a single point of failure which is a risk).

 

IMHO The extra money can best be spent on:

 

1. Twin PDU strips running of two separate phases, get it cabled into the server room (if you have three phase that is, if not, ignore)

 

2. Overspec the UPS's - for two twin processor servers I would suggest two APC UPS 5000 plug one UPS into each PDU, and then on each server plug one PSU into UPS1 and one PSU into UPS2. This way if a UPS fails, or a phase goes down the other will keep the servers going, if all power fails, you should have over 1hour of runtime with that setup. If you want to leverage the runtime, increase the size of both UPS.

 

3. Network Switch – make sure you have a 10Gig core switch for the servers to run off.

 

4. Backup server with secondary NAS target. Replication in hyperV is not a replacement for traditional backup. I recommend a £2k ish server set up in the server room running on 10gig, chock full of SAS drives (SSD preferable if you have the cash) running Veeam Backup and recovery. This doesn’t need to have ginormous storage as it is only a staging server for fast backup and restores, it then automatically replicates the backups onto youre remote NAS chock full of big SATA drives in Raid 6 which stores the long term, weekly, monthly, yearly backups. I use a 12 bay Synology, this can be 1GB Ethernet only as the replication will occur out of hours anyway so speed isn’t a massive issue.

 

5. Air conditioning. Make sure you’ve got air con, preferably, two fixed wall mounted units run from each of you’re PDUs so that if one phase goes down the other air con will keep going. I’ve only got one unit due to budget but if I had a spare few k I’d get two new ones wired in as above.

 

That should easily take care of £20k with software licences and any electrical work.

 

Cheers

Edited January 5, 2016 by PotNoodleTech

[twin--turbo]

I could easily spend 20K. What I would try and squeeze in.

 

2x Physical hosts booting vmware from SDCARD ( Dell 730XD's are good for this )

1x SAN/NAS that can do NFS ( We now use this ove ISCSI as it's faster, more flexible and easier!)

Idealy have 10Gb to the the SAN and Hosts.

Idealy have this duplicated to two switches with something like "IRF"

 

 

We recently got quoted 10K for Dell R730's with 196Gb of ram a couple of spinning disks and 2x18Core XEONS. Just one of these could easily host 36 general server VM's without issue. But we are using them for Citrix Xenapp.

 

TT

[Eudora931]

Subscribe

[dapaulio]

What would you get?

 

I'm trying to get with three decent spec'd HP servers to run Hyper-V instances. Two in use, one as a hot migrate to. All connected to a storage array where the VMs live.

 

Would you do this? If you would what would you choose? Assume 4/5 VMs per host.

 

Not that I have to make this a reality in .... 3 weeks!

 

Over the last two years (ish) we have been overhauling our server room

we have got 3 x DL380 G7, G8 and G9 with 2 x 8 core 90-110GB RAM and 4 port HBA card in each connected multipath to MSA2040 san with 10TB of storage setup in 3 RAID 5 with hot spare volumes (with spare bays to play with)

18 VM's spread evenly across all 3 hosts with plenty of redundancy for failover

All servers and switches connected back to a HP 5406Rzl2, redundant management, 4 x modular 20 gig-t + 4 SFP switch and 1 modular 24 sfp switch (+1 spare module bay for expansion)

 

1Gig fibre to all our edge switches (planning for 10Gig fibre in the works to future proof - doubt implementation will be a while off yet)

 

our core and servers are good. just have to work on the edge switches now. all running ok but are getting old and have a few cascading room switches which I want to pull back to the nearest edge switch cab.

 

95% of computers are within 5 years of age on an active 5 to 6 year rolling upgrade plan

[Davit2005]

I'd stick with SAS personally, SSD is great if you need the IOPS but I'm assuming you don't by the sounds of it. Guess how long do you want to future proof it for?

 

Also in regards to DAS you'd just have both servers with storage, then replicate across, no need for shared storage unless you're doing cluster? Which then would make the third-hot one pointless as you'd either have all 3 live, or wouldn't need the 3rd etc

 

Steve

 

TBH We used to run 105 VM's on SAS based HP EVA storage (connected via multiple FC) without an issue, lots of SQL servers, Exchange, file servers, etc. and was not slow.

 

I don't think Enterprise SSD's are cheap and we are only now using small amount of flash in our EMC.

[dhicks]

I've been struggling with this for months now. They've given me a huge budget, just under £20k for the new secondary site. Which will house 100 kids max as KS3/4 age. As we're a special we really do not use IT in the same way a mainstream does. No video editing, no media work, mostly Office, internet and some locally installed apps.

 

You could skip the whole locally-installed hardware setup and go for remote servers connected by a fiber Internet connection - let it be someone else's problem to take care of the hardware for you.

[Blue_Cookeh]

I'd still vote for three virtual hosts with local storage. Forget "Enterprise" SSDs and get say 6 large "standard" SSDs RAIDed in each system, create a Hyper-V cluster, setup Veeam to back the VMs up to a NAS (I like Synology), and then ship a copy off to Amazon Glacier if your connection is fast enough.

[sted]

tbh ive never seen the advantage of using iscsi devices as vm storage rather than using the internal drives. Surely its still a single point of failure youve just moved it from the server itself to another box meaning that if 1 of 3 (server/san/network switch) things fail you have a problem? Surely its more sensible to have the vhd replicated between hosts to internal drives (as lets face it hdds are cheap shoving a few extra tb in a server isnt much of an issue) and it a simpler setup

 

im willing to be talked round but it does seem an over complicated solution to a problem (again this may be because im used to hyper v rather than vmware etc)

[john]

tbh ive never seen the advantage of using iscsi devices as vm storage rather than using the internal drives. Surely its still a single point of failure youve just moved it from the server itself to another box meaning that if 1 of 3 (server/san/network switch) things fail you have a problem? Surely its more sensible to have the vhd replicated between hosts to internal drives (as lets face it hdds are cheap shoving a few extra tb in a server isnt much of an issue) and it a simpler setup

 

im willing to be talked round but it does seem an over complicated solution to a problem (again this may be because im used to hyper v rather than vmware etc)

 

So a quick one back to your point... I have 0.5PB of storage attached to my VMware cluster how can I fit that with SSD, SAS, SATA and EMC Fast Cache in those hosts and share them please with your method of only have disks in hosts, should say due to rack space needs my hosts are Blades and they only have 2x HDD Slots, even if I swapped them out for 3 series HP DLs they would only have about 10 disk slots at the most....

[Steve21]

It depends on scale of network really. Small networks it's overcomplicating it for "little" reason.

 

Large ones it's 100% needed really unless you're wasting millions on local storage

 

Steve

[Davit2005]

tbh ive never seen the advantage of using iscsi devices as vm storage rather than using the internal drives. Surely its still a single point of failure youve just moved it from the server itself to another box meaning that if 1 of 3 (server/san/network switch) things fail you have a problem? Surely its more sensible to have the vhd replicated between hosts to internal drives (as lets face it hdds are cheap shoving a few extra tb in a server isnt much of an issue) and it a simpler setup

 

im willing to be talked round but it does seem an over complicated solution to a problem (again this may be because im used to hyper v rather than vmware etc)

 

We use multiple connectivity for our setup (multiple SAN switches, controllers, server links, HBA cards, PSU's etc.) , but as we are a college and have 105 VM's probably a bit different scenario than most school setups, we are currently only using 50% of our 92TB available on the EMC.

 

Setup in Hyper-V I guess would be similar for our sort of usage.

[sted]

So a quick one back to your point... I have 0.5PB of storage attached to my VMware cluster how can I fit that with SSD, SAS, SATA and EMC Fast Cache in those hosts and share them please with your method of only have disks in hosts, should say due to rack space needs my hosts are Blades and they only have 2x HDD Slots, even if I swapped them out for 3 series HP DLs they would only have about 10 disk slots at the most....

 

cant you get blades that allow connection to an external source of hdds (ie a blade chasis with no processor just full of disks) ie PowerVault MD3 Serial Attached SCSI Storage Series | Dell UK though i suppose thats just a san via a proprietry cable at least it takes the network out of the loop so its basically just a drive bay add on for a server

 

Granted ive not used them the closest ive got is a setup with 2 rack mount hosts but again internal storage and no replication

 

again im primary school so dont have the data requirements of a secondary but i also dont have an server with less than 3 drives in and usually another 5/7 bays i can use if needed without needing a san etc

[PotNoodleTech]

As DAvit and Steve mentioned - Central storage becomes perfect if you duplicate so there is no single point of failure, two enclosures, two switch's, multiple connections, multiple HBA cards in each host server. Perfect. But very expensive.

 

Most schools can't stump up the budget for that so can't get rid of the singe point of failure.

 

The question has to be asked - is the significant extra cost worth spending to get the extra 0.1% uptime? IN the ideal world, yes. In the real world, probably not.

 

Another good way of doing central storage on a low budget for up to 4 physical hosts whilst keeping high performance is to use a 12gig SAS JBOD array "tray" and a 12 gig SAS HBA card in each host. The tray I still a single point of failure, but theoretically should be less chance than a SAN as the tray is just some disk trays and SATA connections in a box, no complicated software, no nasty risky updates, etc.

[ataylor]

Out of curiosity what SSD's would you guys recommend for servers?

[mavhc]

First question to ask is how much money you'll be getting in the future, £20k now and £0 for the next 10 years is bad, better to only spend £10k and keep the rest for later. Also spending £20k on servers and not having the network to use them well, or the client machines to use them well, is a waste of money.