Storage Advice for a Dell C1100

[tycoonbob]

So about a week ago I picked up a Dell C1100 for a great deal. Has dual quad core Xeons (with HT) and 36GB of RAM, which I will be upping to 48GB of RAM (DDR3 ECC Registered). This box is currently, and will continue, to be used as a Hyper-V host here at home. It's running Server 2012 Datacenter.

Currently, I have a 160GB drive (7200RPM) running as the boot drive, and 2 3TB Toshiba DT01ACA300s running in Storage Spaces with a Mirrored Space. Performance is slightly better than just a single drive, and I currently have 9 VMs running on this box. Disk IO is currently fine, but 6 of these VMs have a SQL Server 2012 SP1 instance running on them, so it's a matter of time before I start to feel a lack in disk IO. For example, if I create a new fixed VHD (all my VHDs are fixed), disk IO for my running VMs come to a halt, but that down time is fine (as it's only a few minutes).

I have 2 other Hyper-V hosts (white box with 8 core FX-8120s and 32GB of RAM -- 2 60GB SSDs in RAID 1 for the OS drive, and a single 1TB drive for VM storage), and I also have a NUS (Network Unified Storage) server. My NUS currently has 4 2TB 7K3000 in a RAID 5, but I am in the process of buying more Toshiba DT01ACA300s to start a RAID 60, which is primarily for write once, ready many kind of data (media mostly, but personal docs, RAW images, home videos, etc). Once the RAID 60 is going, the 4 2TB Hitachi's will be dedicate to my on-site backup, along with a pair of externals, along with AltDrive for full cloud archiving. I currently have two iSCSI targets from this box, witch each target having a single VHD. Build log for my NUS can be found here.

**Please no "RAID is not a backup" preaching, that's not what this thread is about.

Now with the Dell C1100, I plan to use local storage for the majority of my VMs, but a few will run via iSCSI from my NUS (NUS has 4 gigabit NICs, with the Dell C1100 having 2 gigabit NICs -- will be using LACP). My C1100 is the model with 4 3.5" bays, and I am trying to come up with an appropriate local storage scenario. What I have now works fine, but I have a bit of resources (16GB of RAM free currently, and CPU is still under-utilized) that are not being used and disk IO will become my bottleneck. This server does not have a RAID controller, but it has one free PCIe slot for one if I wanted. Right now I am debating doing one of the following:

Scenario 1:
Buy a PERC 6i (or similar), get 4 500GB SAS drives (10k or 15k RPM) and do a RAID 10.
Will yield ~1TB of local storage, which will have a 30GB partition for the OS, and the rest is for local VM storage.

Scenario 2:
Buy a PERC 6i, get 4 1TB 7200RPM drives and do a RAID 10.
Will yield ~2TB of local storage, which will have a 30GB partition for the OS, and the rest is for local VM storage.

Scenario 3:
Buy a PERC 6i, get a 60GB SSD for the OS, and 3 500GB SAS drives (10k or 15k RPM) and do a RAID 5 with the SAS drives,
Will yield ~1TB of VM storage (please no comments on using SSDs as server boot drives)

Scenario 4:
Use onboard SATA (no RAID), buy a Samsung 840 500GB (VM storage for high Disk IO VMs), buy a 1TB 7200 RPM drive and partition 30GB for OS and use remaining for low Disk IO VMs (such as Windows 7/8 clients, VOIP server, etc).

With my current 3 Hyper-V hosts, I have 23 VMs and will eventually have around 32+. My plan is to decommission my 2 white box servers, get another C1100 and have both C1100s identical, and use Hyper-V Replica to replicate my important VMs (DCs, media server, etc) while using System Center 2012 Data Protection Manager to backup ALL VMs to my RAID 60 on my storage box.

The 9 VMs currently running on the C1100 are using around 600GB of storage, so I want at least 1TB-ish for VM storage, but non-priority VMs can run on single 7200RPM disks or iSCSI without a problem.

What are your thoughts on my VM storage situation? Is there another scenario I should be looking at but didn't think of? Any questions or something I left out?

Thanks!

 

[hotcrandel]

I am asking the obvious question only since you haven't bothered mentioning it, Have you considered a dedicated ZFS based iSCSI or NFS storage server to do your virtual machine disk IO?

You can get good effect for fairly cheap hardware, possibly that you have laying around already, and with some disks and a couple SSD's you can have tiered storage that maybe have a total performance value much higher than individual pieces.

Since it sounds like you already have a switch capable of LACP assuming you have free ports you can break the gigabit bottleneck there, too.

The perc setups will work fine, especially if you use SAS disks as well.

 

[tycoonbob]

While I have nothing against ZFS, I have no need for a second dedicated storage box. My NUS is speced as follows:
Norco RPC-4224
Xeon 1220-v2
8GB DDR3 ECC RAM
LSI MegaRAID 9261-8i Raid Controller
HP SAS Expander

Will soon start an 8 drive RAID 60, and expand that out (ideally to 16-20 drives, plus 2-4 warm spares), but currently have the 4 2TB RAID 5 in that. For the majority of my VMs, I don't plan to run iSCSI since my storage box would be a single point of failure for my entire home domain. I will have a few iSCSI Targets running though, for certain VMs that need additional storage or for Windows clients for my test VDI.

That is the reason I am looking for the optimal local storage solution. Which one of my scenarios will provide the best performance for the price it would cost? My guess is the SSD will be the cheapest, and highest IOPS but maybe not the highest seq. R/W.

 

[cafcwest]

I love seeing the Hyper-V love!

I am pretty familiar with the C1100/C2100/C6100 lineup. Based on what you currently own and are looking to acheive, I'd be inclined to do the following:

C1100 - wait for the Micron/Crucial M500 960GB SSD to drop in the next few weeks (CES announcement said Q1 release, $600 retail pricing). Pair with with a 60/64GB SSD of choice for Hyper-V OS install (since you are using System Center for management, are you going to run Hyper-V Server as your hypervisor, or are you using Server/Server Core?). This gives you the space you need for storage, and the IOPS to run the workload.

Use your whitebox or "NUS" as your backup/storage target/whatever. As much as I hate mixing drives, Toshina DT01ACA300's and Hitachi's 7K3000 do play nicely together in RAID (as they are pretty much the same drive). With more than 4 drives, you'd have sufficient performance to present a usable iSCSI target to your VM's for whatever needs you have there.

 

[tycoonbob]

You sound familiar with my world! Yes, I am using VMM to manage my VMs (as well as ConfigMgr for managing updates/SCEP, and DPM for backups). I am running Server 2012 Datacenter on my Hyper-V hosts, and not the Hyper-V server.

$600 is probably a little rich for my blood, but for a 960GB SSD that's an excellent price. I may have to think about it a little more.

I won't be mixing the Toshiba and Hitachi's in the same array. I will eventually have a RAID 60 with only the Toshiba DT01ACA300s, and a separate mSAS or eSATA JBOD (aka, DAS) device with the 7K3000s, used purely for backups. My NUS will be an iSCSI Target, but for for only a few VMs. I am really thinking that the best plan will be to use a single SSD for my disk intensive VMs.