Servers

Hyper-V Host Clustering via ISCSI

Ryan Femling — Wed, 10 Dec 2008 01:06:31 GMT

There's a lot of articles out there about Hyper-V host clustering. But there isn't much about the associated caveats. The main issue I ran into was with creating new physical disk resources. When you create a cluster and add nodes to it, you may end up adding a disk resource from a machine that does not own the cluster group. Yes, contrary to what it looks like in the GUI, there are still cluster groups, and much of the underlying clustering administration is unchanged from server 2003. If you're in a situation where you need to add physical disk resources and the GUI won't see them, here's what you do:

1. On one of the nodes that is a possible owner for the resource, open two consoles as an administrator.

2. Log on to the LUN via the ISCSI administrator

3. In storage management, bring the disk online, initialize the disk, and create a simple volume (note, you can initialize the disk via MBR or GPT, but the disk MUST be a basic disk).

4. In the first command prompt you have open, type diskpart. Type select disk <appropriate disk number here>. And then type detail disk. You should see something like this:
MSFT Virtual HD SCSI Disk Device
Disk ID: 2E3EA1FE
Type : iSCSI
Bus : 0
Target : 10
LUN ID : 0
Read-only : No
Boot Disk : No
Pagefile Disk : No
Hibernation File Disk : No
Crashdump Disk : No

5. In the second command prompt, type Cluster res "<resource name here>" /create /group:"Available Storage" /type:"Physical Disk" This will create the resource.

6. In the second command prompt, type Cluster res "<resource name here>" /priv DiskSignature="0x<DiskID from first command prompt>" The 0x before the disk ID listed is important, and you won't be able to bring the disk online if you leave it out.

7. In the second command prompt, type Cluster res "<resource name here>" /on This will bring your new disk resource online.

When doing this, make sure that the Available Storage resource group is owned by a machine with access to the LUN you're trying to bring online. It seems that Microsoft wants you to grant access to every node in your cluster, but if you're like me and are a bit nitpicky about which machines have access to which LUNs, you'll want to follow these steps.

Name brand VS white box

Ryan Femling — Wed, 12 Nov 2008 23:30:27 GMT

It's interesting to talk to people in this business and find the religious fanaticism that revolves around the support of different hardware vendors. I've met systems people who have variously sworn by HP, Dell, and IBM. But what about the good old white box? If you look at it closely, a lot of businesses are throwing away good money for name brands when a white box machine could be built with the exact same chipset for a lot cheaper.

For example, we do a LOT of work with VMs here at 3Sharp. I built us a VM server with 2 quad-core Xeon procs (2.2Ghz), 32GB of RAM, in a 1u form factor with mirrored 500GB SATA 2 drives. This server cost about $1750 to build, a server from Dell with the same chipset, procs, RAM, and HDD costs $6911, and if I were to switch it out with Opteron procs, the price would still be $5196.

The objections to this approach to hardware generally tend to revolve around convenience and support. My take on it is that if you can save 40-70% on your hardware costs, you can then easily afford to have spare parts for every system you have on hand. That also means that you can have a better response time than the 4-hour onsite.

There are a few keys to making this work though. First, you have to have at least one IT staff member who REALLY knows his hardware. This means knowing when a 20-pin power supply will be able to power a 24-pin motherboard and not void the warranty (the answer is that if you're not going to be putting in a lot of add-in pci or pci-x cards, you should be ok). Second, pick a standard. If you're whiteboxing your servers, that means you settle on one or two motherboards to be your standards. The less variety you have, the fewer spare parts you'll need to keep around. Third, pick your vendor very carefully. The best price is definitely not always the best deal.

Other keys to success with whiteboxing include using a good system monitoring solution. We use Microsoft SCOM, but I've also used Nagios with good success. I've also seen comparable reliability to brand name servers with the units I've built here.

While it may not be practical for large organizations to use whiteboxing, small to medium sized businesses can benefit greatly from it. Their IT staff will be much more intimately familliar with the hardware, and costs will be reduced.

More clustering fun....

Ryan Femling — Sat, 03 May 2008 00:07:52 GMT

As you may or may not know, I love clustering. I've been working with MSCS for a very long time, and I look at it as a way to make my daily life easier. An admin with a cluster can have downtime on hardware in the middle of the day, can do patches at any time knowing that if there's a problem, all you have to do is fail back to an unpatched node. That and it makes for some really cool toys.

Up to the present. I decided to check out MSCS (now known as failover clustering) on Windows Server 2008. I'm a big fan of building your own ISCSI devices, and I like Openfiler and FreeNAS. Both provide a large list of features, and the capability of Openfiler to run a two-node failover cluster of it's own (thus providing redundant storage for your cluster) to be a wonderful thing. So, you can imagine my dismay when I discovered that faliover clustering in Windows Server 2008 requires support for persistent reservations by your iSCSI target. Unfortunately, none of the free ISCSI target solutions support this yet. When they do, I'll be sure to update with more info.

Back to a blast from the past, clustering on windows server 2003. Here's another caveat for everyone out there: We'll often build our nodes with two disks in them and mirror those disks. Be careful about what method you use, some onboard RAID solutions will use BUS0 for their volumes. If this is the case with your RAID controller, switch to software mirroring. The reason is that the ISCSI initiator also latches on to BUS0, putting it on the same bus as your boot volume. This will cause cluster creation to pop up a warning about being unable to find a quorum device (BEWARE, no warning will pop up if you use the advanced creation option, you just won't be able to add any shared disks later).

Hyper-V and VLANs

Ryan Femling — Wed, 23 Apr 2008 18:06:49 GMT

I was excited when I heard that Hyper-V was going to include support for VLANs. What I didn't realize was that there would be some caveats to the deal. If you plan on using tagged VLANs to get support for multiple networks from one physical port, you need to know that Hyper-V works on a one-to-one relationship with interfaces.

To make it all work, you need to create your virtual interfaces (per the software and driver for your NIC) and assign them to the appropriate VLAN. Don't bother putting in any static IP information because as soon as you create a network in Hyper-V to accomodate your VMs, Hyper-V will create another connection on top of the VLAN adapter you've already created. Once this new interface is created, you can enter any static IP addresses there. Additionally, if you don't take into account the extra caveats below, you will have no network connectivity.

EXTRA CAVEATS:
1. Make sure you specify the correct VLAN number in the virtual network.

2. Make sure you specify the correct VLAN number in the network settings for the virtual machine.

Microsoft's UC Launch

Ryan Femling — Fri, 19 Oct 2007 17:48:00 GMT

As you may or may not know, 3Sharp played a part in the UC launch. The keynote demo given by Eric Swift was implemented and supported by Peter Kelly and myself. The launch event itself was amazing, and if you ever get a chance to attend one I highly recommend it. A few notes on the keynote demo though: The audio quality of the webcast didn't do justice to the capabilities of the voice portions of the demo. Voice quality over all the attached devices (including and especially Round Table) is excellent. It does not, however, benefit from piping over a PA. Outlook voice access appeared to have a hard time understanding Eric's commands. This only happened when he was hooked up to pipe the sound over the house audio. I use OVA on a regular basis here at 3Sharp, and never have those problems.

The big point I'm trying to make is that these features work exceedingly well in the real world, and the sound quality is top-notch whenever you add voice to the mix. Watch the demo and pay attention to the features being shown, and remember that when you use them the quality of voice is likely to meet or exceed that of your PBX.

The keynote can be found here for those who missed it: http://wm.istreamplanet.com/customers/ms/10162007_300k.asx

An article on high availability for SharePoint 2007

Ryan Femling — Wed, 15 Aug 2007 19:19:00 GMT

There's an article I wrote in this month's Windows IT Pro magazine that provides an overview of high availability solutions for MOSS. Part of the article is available here: http://www.windowsitpro.com/Articles/ArticleID/96301/96301.html?Ad=1 The article is brief and takes a very high level view of architecting for high availability. Over the next few months, I'll try to post in-depth on the subject, breaking it down into the different components.