This is entirely a rant, and only tangentially instructional. It is related to the virtualization malarkey I’m doing at the moment, but won’t actually help anyone set up a similar infrastructure. It might save someone a bit of brain-ache in the future, but mostly, I’m just annoyed.
To recap briefly: I want shiny new virtual servers in a high-availability cluster, and to do that, I have 2 beefy DL380 G7s and a single DL160 G6 to act as a physical DC and administer the VMs. There’s an LTO5 tape autoloader for backups that connects by SAS cable.
My plan was for the DL160 to be running Systems Center Virtual Machine Manager (SCVMM) to manage the DL380s, and to run Systems Center Data Protection Manager (SCDPM) for the backups. Not unreasonable, right?
SCVMM, to work fully, has to integrate with Systems Center Operations Manager (SCOM) which enables the Performance Resource Optimisation (PRO) that intelligently balances the load on your virtual hosts to maximise performance. This is functionality that I desire.
Unfortunately, SCDPM cannot co-exist on a server with SCOM. Why, I don’t know – I’ve genuinely not seen an explanation, just that it can’t be done.
My only hope is that SCOM can run in a VM and use the SCVMM SQL instance remotely, because to further complicate things, my SCVMM license includes an instance of SQL, my SCCM license includes an instance of SQL, but SCOM doesn’t. So SCOM has to piggyback on one of those SQLs, which I imagine will need to be the SCVMM instance in order to actually integrate. As far as I’m aware, though, SCOM is heavy on the database traffic, so this isn’t ideal. Please feel free to enlighten me on this matter if you have actual knowledge of this daftness.
At this point my head started to hurt a bit so I decided to tackle another potential problem – the DL160’s connection to the SAN. This is now a bit of a rant about HP and their form factors as well.
As it stands, the SAN connects to the fibre switches and the DL380s connect to the same switches. The DL160 does not, which I’m concerned will give me problems with setting up Cluster Shared Volumes in SCVMM and will slow the backups to the autoloader down. No problem – I can just connect the DL160 to the fibre switches as well. Sorted.
A glance at the shops tells me I need a spare PCI-E x8 slot for a fibre HBA card. Looking inside the DL160 (as HP Systems Management insisted everything was empty, a blatant lie), I can see the x16 slot occupied by the P212 SAS controller x8 card (with external port for the autoloader), an x8 slot inside and two x4 slots. It might take a bit of kludging, but surely I can use that x8 slot and run a cable inside, or use the x4 slots – they shouldn’t limit the speed that much if I can get the right riser.
Unfortunately, the x8 slot is reserved solely for another controller, the P410. It literally cannot be used for anything else. How about the x4 slots then? HP don’t sell risers for them. Any risers. At all.
What is the point in providing expansion slots that cannot ever be used?
Fine, so, fibre HBA and SAS card can’t live together in the DL160. Maybe I can move the SAS card to one of the DL380s, that do have spare slots, and then I can use the x16 slot of the DL160 for a HBA.
At this point SCDPM rears its awkward head again. It cannot be installed on the node of a cluster, i.e. one half of a HA pair. Maybe I can run DPM in a VM then – but Hyper-V cannot expose external SAS connectors.
So, I have three options:
1) Hope that the DL160 doesn’t need the fibre connection, that it can talk to the SAN through the IP network and administer the CSV and everything that way, and that the backups are quick enough over the IP network.
2) Buy an FC autoloader, and connect everything via the fibre switches. Cost: about £4k and a wasted £4k on the other autoloader. Not really an option at this point.
3) Buy another 2U form factor server, plug the SAS card and a fibre HBA into it, and run SCDPM on that.
The third option would solve the problems between SCOM and SCDPM, but it would still cost a small fortune due to the price of fibre HBAs (£1.3k for a dual port card; even scrapping redundancy as it’s only backups, still £800 for a single port card. I need another card for the DL160 as well). More to the point, it would mean that, in a new virtual environment, where I am attempting to maximise the use of resources and minimise the physical infrastructure, using technologies designed for and sold on that promise, I still have to install a single purpose server to spend most of its time sitting idle and only occasionally actually do something.
With enough sitting down beforehand, perhaps I could’ve predicted this and gone with option 2, but it’s difficult to predict these foibles until everything is in front of you and you can physically work everything out.
Alternatively, Microsoft could learn to actually integrate the System Center suite, instead of just branding disparate products under one name for the sake of marketing.