Educational Enterprise Server Build

[dude_56013]

(Scroll to the bottom for a TL;DR)

Hey all, I'm not sure if you read my long update in RCC, but I will likely be building a server in the coming months. I've spent many hours researching what the server components will need to be, as I'm fairly well well-versed in custom desktops, but have virtually no experience in server builds in terms of how far certain hardware can be pushed for a particular number of server users.

The organization that I'll be building the server for is a special education facility that deals with educating students who do not fit into ordinary classrooms, either because of emotional or behavioral needs (such as those students who may be on the spectrum, or those with EBD, etc.). The school has somewhere between 15-20 FTE's and educators, with around 80 students.

The request for a server in this building is because the organization would like staff members to begin housing documents on a server instead of Dropbox (or, God forbid, their physical machines with no backup). They would also like the students to begin using the servers to log into lab computers so that they can save and access files.

Unfortunately, due to the nature of this organization, Google Apps for students is out. That is not a route that they want to go at this time. As students begin their higher grade-level education in this facility, it becomes more about job preparation skills. This means that the students need to be experienced in MS Office and other professional programs, not in Google Apps/Docs which really haven't hit mainstream business use at this time (though I love using it myself).

Here's what will likely be occurring on the server. It'll be running ESXi host. I haven't done the math to see how many VMs will be needed yet, but here's an idea of the roles it will have:
-DC1 (AD/DHCP/DNS)
-DC2 (replication)
-File Server
-App Server (used to push out updates/images, AntiVirus, etc.)
-Vcenter Server (to manage the other VMs from console)
-Veaam (or some kind of backup) server

Essentially it's really only going to get “pushed” for AD/DHCP/DNS requests, along with the file server aspect of it. The app server will not be a constant drain on the server, nor will Vcenter or Veeam. DC2 will be used for replication (I guess technically all DC's are now primary and replicate) if something should go awry on DC1 (such a MS update that breaks something, etc.).


----HARDWARE----
The base hardware that I've picked for this setup looks like this:
Chassis Barebones = ASUS SUPERMICRO SYS-6028R-WTRT 2U (LGA 2011/DDR4): SUPERMICRO SYS-6028R-WTRT 2U Rackmount Server Barebone Dual LGA 2011 Intel C612 DDR4 2133/1866/1600 - Newegg.com

CPU = Intel Xeon E5-2630 x1 (with ability to expand to dual CPUs in the future, if needed): Intel Xeon E5-2630 v3 Haswell-EP 2.4 GHz 8 x 256KB L2 Cache 20MB L3 Cache LGA 2011-3 85W BX80644E52630V3 Server Processor - Newegg.com

RAM = 64GB DDR4: SAMSUNG 16GB 288-Pin DDR4 SDRAM ECC Registered DDR4 2133 (PC4 17000) Server Memory Model M393A2G40DB0-CPB - Newegg.com

Backplane cables w/ SGPIO for the chassis: http://www.amazon.com/Supermicro-CB...UTF8&qid=1451921010&sr=1-1&keywords=CBL-0188L

RAID AoC = LSI MegaRAID SAS 9271-8i: Robot Check
--
That base config, as is, sits at about $3,300 (without storage or software—discussed below)


----STORAGE----
That leads me into my next question, and that is about storage options. I would really like to see them go with an SSD storage option. I know it seems overkill, but their IT budget for the person helping them out (which is me right now) is only FOUR hours per WEEK. That means this thing needs to be reliable. I just am leery of traditional drives dropping all the time and having to rebuild the raid array. I know maybe I'm paranoid, but I've just heard horror stories. I know that SSDs still experience failure, but my hope is that it would be less than a traditional drive.

I've got over and over different storage options and pricing, and these are the main two that I came up with:


*Traditional 3.5” 7.2k SAS HDDs in RAID10 configuration (please don't try to sway me here, I've done about four hours of research on this, and with the rebuild times on larger drives, this is what I've chosen. A daily image of the server will need to be kept (hence the need for Veaam) in case more than one drive drops and the array fails before the hot spare rebuilds.)

It will be a 4TB array with 4 of these guys (Seagate Constellation ES.3 ST2000NM0023 2 TB 3.5" Internal Hard Drive - Newegg.com) plus a hot spare.

Since these are traditional drives, I'll probably enable write-back on the RAID controller, and thus will need a battery to control that write-back during any power issues on the machine. (LSI LSI00279 MegaRAID LSIiBBU09 Battery Backup Unit for MegaRAID 9265 and 9285 Series--Avago Technologies - Newegg.com)
--
Total cost for this storage option: $920.00 // Total hardware build cost: $4,220.00


*Traditional 2.5” 10k SAS HDDs in RAID:
2.4TB array with 4 of these guys: (Seagate ST1200MM0158 1.2TB 10000 RPM 128MB Cache SAS 12Gb/s 2.5" Internal Hard Drive - Newegg.com) plus a hot spare.
RAID controller will again need a battery.
2.5” drives will need tray adapters (SUPERMICRO MCP-220-00043-0N Hard Drive Tray - Newegg.com)
--
Total cost for this storage option: $2,400.00 // Total hardware build cost: $5,700.00


*Enterprise SAS SSDs in RAID10:
2.4TB array with 4 of these guys: (http://www.newegg.com/Product/Product.aspx?Item=N82E16820167353) plus a hot spare
Raid controller will not need a battery due to these being SSDs and that is not a recommended setup.
Will again need drive tray adapters
--
Total cost for this storage option: $4,000.00 // Total hardware build cost: $7,300.00


----SOFTWARE---
Not much to say here. I will need to get quotes on licensing form a MS Vol. Licensing distributor, along with Veeam and VMWare quotes.


**TL;DR**
Will the hardware above be sufficient for the roles/# of VMs listed for a school with ~20 FTEs plus 80 students?
What storage option do I need to go with? Can I get away with non-SSD for file server purposes, or do I need to make the case for SSDs?
Do my software needs seem appropriate? I'll need to get vendor quotes on the volume licensing, backup solution, and VMware solution.

Thanks for any input.
dude

 

[dude_56013]

Bumping to see if any of you have any expertise on the server side that could thumbs up or down this, at the very least. Thanks, all!

Sent from my Nexus 5X using Computer Forums mobile app

 

[SethWilliams]

I'll take a stab at it for you. I work for my local government in IT, and we recently went through a similar situation. We ended up buying a Lenovo Server and then the next year bought the same model as a backup/redundant server. Here's what we spent ~2 years ago, which I adopted to your specs.

Lenovo RD650 E5-2620v3
RAID Adapter
2 x SSD for OS
6 x SAS HDD for Storage
Redundant Power Supply
3 Year Warranty
64 GB RAM
Total Cost: $4,800

We have our SSDs in RAID 1 for the OS and main programs. The HDDs are in RAID 6, which I highly recommend. Why are you so convinced that RAID 10 is the way to go? I would be interested to hear your reasoning, as I have setup multiple servers here that I have used other levels of RAIDs on for different reasons.

Since we have two of the same server, our domain controllers are on different servers. I don't see a lot (though there are a few) benefits of putting two domain controllers on the same virtual host. I don't know what your total budget is, but I would highly recommend looking at getting two servers, even if one is not as powerful as the other. You could cut out the HDDs as well to save some money. This would give you redundancy as well as the ability to spin up a VM on another machine if one host goes down, and let me tell you from personal experience, this is huge. I hate getting the phone calls asking how long it's going to be before our software is back up. If you do not have the budget right now, have them plan on buying another server next year for a backup.

AD/DHCP/DNS will not push your server. We have our domain controller (as well as other roles) setup as a quad core, 12GB RAM, and it has more than enough power. We run Veeam as well, and it is not resource intensive at all. Make sure you have your backups stored on another server as well. Something like AWS may come in handy for you.

I'm not sure where you are located, but check on your state contract prices. TN has contracts for each type of item, and typically I can get the best prices based on that contract. I would definitely recommend getting a name brand server WITH A WARRANTY. What happens next year when you custom built this server and you are no longer working for them and somebody else comes in to work on it? Make sure to comparison shop as well.

Hope that helps!

 

[dude_56013]

I'll take a stab at it for you. I work for my local government in IT, and we recently went through a similar situation. We ended up buying a Lenovo Server and then the next year bought the same model as a backup/redundant server. Here's what we spent ~2 years ago, which I adopted to your specs.

Lenovo RD650 E5-2620v3
RAID Adapter
2 x SSD for OS
6 x SAS HDD for Storage
Redundant Power Supply
3 Year Warranty
64 GB RAM
Total Cost: $4,800

We have our SSDs in RAID 1 for the OS and main programs. The HDDs are in RAID 6, which I highly recommend. Why are you so convinced that RAID 10 is the way to go? I would be interested to hear your reasoning, as I have setup multiple servers here that I have used other levels of RAIDs on for different reasons.

Since we have two of the same server, our domain controllers are on different servers. I don't see a lot (though there are a few) benefits of putting two domain controllers on the same virtual host. I don't know what your total budget is, but I would highly recommend looking at getting two servers, even if one is not as powerful as the other. You could cut out the HDDs as well to save some money. This would give you redundancy as well as the ability to spin up a VM on another machine if one host goes down, and let me tell you from personal experience, this is huge. I hate getting the phone calls asking how long it's going to be before our software is back up. If you do not have the budget right now, have them plan on buying another server next year for a backup.

AD/DHCP/DNS will not push your server. We have our domain controller (as well as other roles) setup as a quad core, 12GB RAM, and it has more than enough power. We run Veeam as well, and it is not resource intensive at all. Make sure you have your backups stored on another server as well. Something like AWS may come in handy for you.

I'm not sure where you are located, but check on your state contract prices. TN has contracts for each type of item, and typically I can get the best prices based on that contract. I would definitely recommend getting a name brand server WITH A WARRANTY. What happens next year when you custom built this server and you are no longer working for them and somebody else comes in to work on it? Make sure to comparison shop as well.

Hope that helps!

I appreciate you stopping by and giving your input. I'll try to address some of your question, but can't promise I'll keep them in order.

First off, with a 4-drive array, RAID6 makes zero sense for several reasons. First is the inherent risk with a drive failure, leading to long rebuild times, and the increased taxation on all disks during a rebuild. The next is the obvious performance hit of RAID6 vs. RAID10. Trust me, I used to be in the RAID5/6 camp years ago, and it's never been worth it based on experiences I've had. Everything will be kept backed up off-site anyways, with physical access to this backup for recovery, so RAID10 makes even more sense should anything actually completely fail. Here are a couple of forum articles that might help you understand my reasonings.
Link 1
Link 2

In terms of comparable off-the-shelf machines, I have looked into it. Similar configurations from Dell or Lenovo are falling anywhere between $500-4,000 more expensive. I've spent about 12 hours in spreadsheets comparing machines for this client, as I thought off-the-shelf was the way to go at first as. It's quickly becoming clear that it is not. Not with the upgrades and customization we are looking to do on this server. Everyone knows that with an off-the-shelf unit, as soon as you start ticking boxes for added features or components that are beefier or faster, the costs really start to add up at a rate much quicker than purchasing the upgrades yourself.

The warranty is the one sticking point. However, the server will be sold through my business. I'm 10 miles from the job site. I can be there 24/7/365. Can't say the same for a Dell or Lenovo tech. Yes, they offer the 3 year warranties, but if we're down for twice as long, it makes no sense. I'll likely add a 1-year warranty into the cost of the server proposal to give them a sense of security with this. The only thing that wouldn't be readily replaceable in a failure situation would be the motherboard. Everything else is literally an off-the-shelf component that any technician with common sense could replace.

Right now, two servers is out of the question. Not just simply for the hardware cost, but the software licensing cost of a second CPU w/ Windows Server.

There are other factors at play here, in this budget, such as M/S licensing fees for the server, the CALs, and Office.

Having two domain servers on one host will allow for staggered Windows updates. Should updates break one of the DCs, then the other will provide the failover. Since it's not too taxing and the server licensing is per CPU anyways, it's a pretty easy decision to make. It also provides a way to staggered new MS Server releases. When 2016 comes out later this year, it will provide the ability to upgrade one DC at a time to prevent any issues.

 

[SethWilliams]

Couple of things to note:

With the tasks that you are looking at doing, none of them are going to push the RAID, even if it's in RAID 5 or 6. In RAID 10, once one drive fails, you then have a single point of failure. Even though rebuild times may be faster, if that drive fails, you are up a creek.

Regarding building a machine versus buying off the shelf, you should be able to get cheaper pricing for education than you can by buying off newegg. I personally would not build a server and do not know of anyone in the professional world that has built their own server.

The warranty can be a big deal, IMO. I live a mile from my work but I still want a 3 year warranty with a guaranteed response time. We don't stock extra RAM sticks, HDDs, or processors for each server, so I would rather them overnight me the part than having to drive an hour or more to buy a part. Also, do you never go on vacation? Or never sleep? I know from personal experience that getting those calls in the middle of the night are not fun and can be stressful. Having a warranty is a big selling point for me, and it should be for your client as well. What if they decide to use another company besides you two years from now? That new company will charge them $150+/hour to work on their server. Those costs add up quicker than buying a name brand server with a warranty.

If costs are that big of a factor, why not run linux and get two servers? It's more important to have redundancy with two linux boxes than have one over powered machine running windows. There are plenty of $3,500 servers that are more than powerful that you could buy two of to accomplish your goals.

I have never had Windows updates break one of our DCs. And if an update did break it, I would simply restore the backup and we are back up and running in 30 minutes. You get a new MS server release approximately every 4 years, right? I don't see that being a viable reason to run two domain controllers on one virtual host. If that's what you are worried about, I would spin up server 2016 in a VM and use it there first and then deploy it to your server. I would work on the upgrade on a Friday night or Saturday, then come Monday morning the server should be ready to go.

 

[dude_56013]

Couple of things to note:

With the tasks that you are looking at doing, none of them are going to push the RAID, even if it's in RAID 5 or 6. In RAID 10, once one drive fails, you then have a single point of failure. Even though rebuild times may be faster, if that drive fails, you are up a creek.

On paper, your arguments hold up, I get that mathematically RAID5/6 are more "resilient" than RAID10 in terms of how many drives fail. But in the real world they do not; there's many other factors other than just drive failure probability. I'll be going RAID10, and we'll just leave it at that. We can agree to disagree on this one.

Regarding building a machine versus buying off the shelf, you should be able to get cheaper pricing for education than you can by buying off newegg. I personally would not build a server and do not know of anyone in the professional world that has built their own server.

We've got many companies around here that build servers, and I'm in a pretty rural community. Our local school district just ordered two no-name servers from a local servicer. We're not talking Intel Blade-level quality and performance here. It's not like we're building a data center.

I've check around. We are part of CDWG and get pretty steep educational discounts. Still can't touch a barebones server build with off-the-shelf prices from Newegg/Amazon. I was just as surprised as you probably are hearing that. First thing I checked was CDWG's Lenovo offerings, and it was actually /after/ that when I decided a barebones would be more cost effective.

The warranty can be a big deal, IMO. I live a mile from my work but I still want a 3 year warranty with a guaranteed response time. We don't stock extra RAM sticks, HDDs, or processors for each server, so I would rather them overnight me the part than having to drive an hour or more to buy a part. Also, do you never go on vacation? Or never sleep? I know from personal experience that getting those calls in the middle of the night are not fun and can be stressful. Having a warranty is a big selling point for me, and it should be for your client as well. What if they decide to use another company besides you two years from now? That new company will charge them $150+/hour to work on their server. Those costs add up quicker than buying a name brand server with a warranty.

The thing is, guaranteed response time is going to include a hefty fee above and beyond the simple parts and delivery service that Lenovo and Dell offer as their entry-level warranty. I know that warranties are a pretty valuable and significant part of many companies' technology plans and budgets, but honestly, this place can save their money and use it elsewhere. They are already stretched thin enough as it is.

You either front load this at the cost of several hundreds of dollars to get a 24/7/365 warranty (which does not come standard on any machine) and possibly never use it, or you don't load this in the cost and leave room in your tech budget for possible work in the future. So what if I'm not there for some reason in the next couple of years? You pay someone who has half of a brain $100/hour to work on the thing for a couple of hours.

Trust me, I understand your view. I just got done working IT for a mid-size bank for about the past two years. We had a DR site, multiple physical servers, nearly 100 VMs, etc. We paid up the wazoo for 4 hour response times, overnight parts, warranties, etc. But it was all warranted--when you're in that line of business, you can't mess around with downtime. But this is different. It's an alternative educational school. They'll be happy with a server, period, because they've gone without one for a very long time and have had to deal with significantly greater challenges and issues than a server being down for a day or two could ever pose.

If costs are that big of a factor, why not run linux and get two servers? It's more important to have redundancy with two linux boxes than have one over powered machine running windows. There are plenty of $3,500 servers that are more than powerful that you could buy two of to accomplish your goals.

Linux isn't going to happen. Not in this environment.

I have never had Windows updates break one of our DCs. And if an update did break it, I would simply restore the backup and we are back up and running in 30 minutes. You get a new MS server release approximately every 4 years, right? I don't see that being a viable reason to run two domain controllers on one virtual host. If that's what you are worried about, I would spin up server 2016 in a VM and use it there first and then deploy it to your server. I would work on the upgrade on a Friday night or Saturday, then come Monday morning the server should be ready to go.

Every single place I've been has run multiple DCs on a virtual host. It may not be needed, but it damn sure is a nice peace of mind when something happens to one of them, which has happened. It's only using a bit of hardware from the host and due to MS licensing, you aren't paying anymore to have another DC running. It seems like a no-brainer to me. Another server box would be ideal--you are correct--but you end up bringing in a lot of other factors to the table when you start talking about that (more time for updates, more security needed, more hardware cost, twice the amount of hardware to upkeep and replace, etc. etc.)

At the end of the day, I may back down the specs of the server a bit, but I think this is a pretty solid build. I will be checking with CDWG one more time before anything is finalized.

 

[crazyman143]

I think your build looks solid and I went with a similar setup for a small business I manage IT for. I would agree seth with that having a separate physical box for your second domain controller would be ideal, but it's understandable why you might not do that in this scenario. Just make sure you take regular offsite backups no matter how rock solid your raid is.