Server replacement - how much storage?

[SteveB_NI]

I've never specc'ed a server before, so would really appreciate it if anyone was able to advise / feedback on the following replacement server options.

 

Background:

Number of users: 780 students + 10 staff (students mostly use this network with the addition of some teachers, staff do the majority of their curriculum work - such as SIMS usage etc on another separate managed service network)

 

Current Set Up:

 

2 x Domain Controllers + storage breakdown as follows

 

1. 3 HDDs (500GB each) in a RAID 5 formation - usable space 930.50 GB

• Pupil MyDocument folders - 292GB - 60% free space currently
  
• Staff MyDocument folders - 300GB - 5% free space currently (* to be fair this is mostly my own stuff which I need to clear onto external HDD!)
  
• Deployment / Shared Area Folders - 297GB - 16% free space currently
  

2. 3 HDDs (300GB each) in a RAID 5 formation - usage space 557.75 GB

• WSUS partition - 195GB - 90% free space currently
  
• Staff Shared folders 2 - 195GB - 100% free space currently (not used at all)
  
• Deployment / Shared Area Folders 2 - 128GB - 66% free space currently
  

 

Now, I'm hoping to maintain the high-speed drive choice and go for SAS drives (10K RPM) and hot swappable.

 

I quite like the look of the Dell PowerEdge R730 - Xeon E5-2620V4 2.1 GHz - however, this comes with a 300GB SAS 10K drive. For the 2 servers I would then be proposing:

For DC01 - purchase 5 600GB SAS 12Gbps 10K drives - RAID 6 formation gives me 1800GB storage space.

For DC02 - purchase 1 more 300GB SAS drive, then use the 1 300GB drive which came supplied from DC01, and in RAID 5 formation gives me 900GB storage space.

 

I plan to keep roles the same on both servers for the time being - and the increased storage on DC01 should be adequate, but I'm concerned about needing to add more drives in the future - is this possible without starting totally afresh?

 

Thanks for any advice

[victory2015]

i would be going for virtualisation with your new setup as there isn't a reason not too be using vmware or hyperv this would give more flexibility and increase reduncy etc.. i would also split of the roles a bit more as well across 4 vms.

 

2 vms per server

 

dc01 - just dc functions dhcp, dns etc...

fileserver01 - staff documents / shared area 1

 

dc02 - same just basic dc stuff & WSUS

fileserver02 - student docs / shred area 2

 

this gives you much more flexibility and you are using the server resources more effectively . I would also make sure both servers are identical storage etc..

 

so looking at your option.. id go for

 

Dell PowerEdge R730 - Xeon E5-2620V4 2.1 GHz each with 3 600GB SAS drives giving 1.2TB RAID 5 per server in storage and a 300 GB Sas for hyperv or vmware hypervisor.

Edited May 29, 2017 by victory2015

[snagrat]

I wouldn't do RAID5 with 3 drives - I would always do 4 drives.

 

But if you are doing 4 drives may be worth doing RAID10 for the extra speed

[SteveB_NI]

I wouldn't do RAID5 with 3 drives - I would always do 4 drives.

 

Thanks for both your replies - will definitely look into the Virtualisation - that's a great suggestion on how to split it up. I have no experience with virtualisation so would be interesting to learn!

 

Can I ask why you would suggest at least 4 drives for RAID 5 out of interest?

[AlanD]

I'm sort of surprised how little space you use....perhaps our staff and pupils are more wasteful with space. I have seen a noticeable increase in media files for use in languages and our film club (who have been making short films) adding disproportionately to our requirements. In both of these cases - I would have resorted to some secondary NAS box if necessary. As a guesstimate I would be looking at least 3 times your current storage capacity. Like you - we are not currently virtualised. It practically doubled the cost 3 years ago and we found that too hard to justify - even though we would have liked it. SANs are cheaper now - and hyperV on Server 2016 is so good you don't need to look at VMware. And you can do clever stuff like using SSD/NVMe as "hot" cache and slower but cheaper 7200rpm as cold(er) storage. In a SAN you would get RAID 6 and be able to allocate disk sizes with more flexibility between the servers as your demand grows. You don't say if you have other servers - like MIS, or filtering, (cloud email?). Do you run GADs/GAPs/Azure Connect on existing servers?

 

Consider what you are going to do about backups - capacity and timings (I'm a fan of windows backups - its not pretty but its free) and you will certainly want to ensure this includes some form of "Off line" backup like tapes or cloud.

 

And consider your backbone/core switch. Not much point in a brand new wizzy server if you throttle all the data to your network. Maybe 10G server connections into switch are not unreasonable. We use 4 x 1GB LAG links from each server - which with a 4 disk RAID don't come close to saturating.

[minimoo]

So most of the data is on a separate network managed by a company ? - hence the low space usage? (Do students also have access to that?)

 

I'm assuming that students may also have access to office365/google docs in some form - so that's worth bearing in mind in terms of space.

 

We've used dell servers for years, and recently bought one of their compellent SAN's + 2 vmware boxes in front.

 

As others have said - having DC + other roles on same box isn't great.

 

In terms of RAID5 - I think you might find dell recommend running RAID6 over that these days - i.e. if you are thinking of running 3 disk raid 5 + hot spare, make that a 4 disk raid 6 setup. The "problem" with raid5+hotspares is if a disk drives, you kick off a rebuilt onto the hotspare, and during the raid build the raid controller drops another disk... see

 

One thing to remember - whilst you might like a DC under warranty.. once student/staff data is off - i'd be inclined to keep an extra dc running - i.e. you could run wsus on the existing out-of-warranty dc01 (maybe with a rebuild to a newer verison of windows if the hardware supports it)

[SteveB_NI]

Yes most of the staff storage is on the other managed network and students also have access to this.

 

This network which I look after entirely is used to service the Ict suites, sixth form study and a few random open access areas - it's not used thoughout the rest of the school.

 

However we still need to accommodate all students as they use it at some point in their time at the school.

 

MIS (SIMS) is taken care of on the other network as is the bulk of the printing. I have a basic Win 2k8r2 printer server on our own network too.

 

Great idea to reuse one of the dcs as wsus and even potentially as an additional shared area after it's been decommissioned as a DC.

[minimoo]

What's the rationale for now having a single network/setup?

[SteveB_NI]

We aren't - we are continuing to keep them separate. I just need to refresh our own network servers.

 

It's purely cost related why we have our own network - we simply could not afford the same amount of machines via the managed service contract. All Northern Ireland schools use this contract it's not our choice.

[Norphy]

 

Can I ask why you would suggest at least 4 drives for RAID 5 out of interest?

 

Frankly, with modern drive capacities, you shouldn't be doing RAID 5 at all.

 

The trouble with RAID 5 is one of risk. Should a drive fail, the array needs rebuilding when the drive is replaced or a hotspare takes over. The data is copied from the parity on the remaining drives which increases the activity on them considerably and increases the risk of an unrecoverable read error.

 

Ideally, depending on the application, you should be using RAID 6 or RAID 10 as these are both able to withstand more than one drive failing.

[victory2015]

personally only reason i suggested raid 5 is i assumed you have a budget and was trying balance the servers for virtualization.. if you afford one more drive per server id use raid 6 too but sometimes you have work with what you have.

 

id recommend vmware in your setup u can use the free version initallay but it give you far more flexibility and allow you too expand storage with ease.

[AlanD]

OK. So this network is only for curriculum use by students (and few staff) ? So to some extent its not so "mission critical" as staff use a management network with MIS etc. Yes ideally RAID 6 provides a degree of extra redundancy. Its rare for a disk to fail - but they do fail. RAID 5 covers that. For a second rare failure to happen before the RAID is rebuilt - yes its possible. But surely that is what backups are for. If this "network" is only for curriculum use - I think I would consider just one high performance/proper server, and make the second a DC something cheap with SATA drives with no real role to play except to be able to recover the domain more easily should the first server fail. And I wouldn't bother with a SAN and virtualisation.

[markwilfan]

With raid 5 if you have a dead disk in the array and an Unrecoverable Read Error on a parity sector on another drive it is goodbye data. Raid 6 or 10 it.

[markwilfan]

For the record too I've had 4 dead disks in the last year. All seagate

[Ashfold]

One disk failing is very rare however when you rebuild an array you have to read every byte of parity on the drives to recreate the new drive and if you get one read error the RAID will fail.

 

These errors have occurred whilst you have been using the array but it has been caught at a higher level or only affected one file and has been dismissed by the user or yourself however, during rebuild there is no higher level error correction and the rebuild fails.

 

If you are happy to recover from back up and eat the down time use JBOD it gives you more storage if you don't want the downtime use RAID 6 or 10.

 

There are loads of links explaining the maths but basically if the array is made up of disks bigger than 2TB or more than 7 disks you are likely to have the rebuild fall over.

 

RAID 6 is going the same way because we are building bigger and bigger arrays but the unrecoverable read error rate (URE) of disks is staying the same at between 10^14 10^16 reads so in 2020 we should be using RAID 10 or a new triple parity RAID.

Edited May 30, 2017 by Ashfold
typo

[AlanD]

I'm not arguing against RAID 6 (or 10 - although 10 wouldn't generally allow any two disk failure unless it happens to be a particular pair the disks)...but simply weighing up the odds against the cost of day to restore your disks. If - as in this case - there are no dependent services or other operations performed by the server - building from bare metal would take less than half a day - even including restoring the data. I agree - it would be an annoying waste of time - but as one disk failing is unlikely in 5 years - lets say 1 in 10. Then the chance of two disks failing should be 1 in 100 which is pretty low.

 

If it was mission critical - and ran school office systems, etc then I wouldn't hesitate to recommend a RAID 6 - not least of all because its not a large part of the cost.

[SteveB_NI]

Interesting points, and I thank all of you for taking the time to reply.

 

I initially liked the sound of having a further safety measure in terms of RAID 6 but I agree as it's not totally mission critical I think the additional storage available by using RAID 5 may be the way to go.

 

Long way off even getting these ordered, so thanks for the inputs, very useful!