Blog Posts

I ran across Scott Lowe's blog entry on Melio FS, Hyper-V, and VMware ESX earlier today. Scott talks about his discussion with Jeff Woolsey - Senior Program Manager for Virtualziation at Microsoft - while at Tech Ed. Specifically how Microsoft has allowed storage partners to create a cluster file system that can be used by Windows 2008 \ Hyper-V. Such a cluster file system could enable the VMFS type functionality of having multiple physical hosts access a shared disk at the same time. This would remove the requirement of one LUN for one VM when doing quick migration with Hyper-V. Specifically Sanbolics' Melio FS was identified as capable of doing this.

I think this is excellent news and means that Hyper-V customers will have a cluster file system option for their Hyper-V hosts. This can greatly simplify storage management when you have more than just a few VMs. On the downside is the additional cost of such a cluster file system. There is not pricing listed on the Sanbolic website, but the Network World article states that the price is $5000 per host. If the cost is anywhere near that, then the price of VMware's ESX doesn't seem that bad in comparison to Hyper-V - assuming that you need or want the cluster file system capability.

Todd

Our shipment of four brand new PowerEdge R805 servers arrived yesterday afternoon. We got them loaded up with Quad-Core AMD Opteron 2356 2.3 GHz processors, 64 GB of RAM, and a Quad-Port Intel NIC (in addition to the four onboard NICs). I just love opening up new servers and I was pretty excited to open these up and get them in the rack - but not as excited as this.



Here is a picture of the new servers in our lab from this morning. I have also uploaded a few more pictures in the Photo Gallery that show the inside of the R805, the included VMware ESX Server 3i, and the servers installed in the rack.

The only thing left to do now is use VMotion to get some VMs onto these systems and put them to work.

Todd

What is the best server for virtualization? This is a question that comes up often in discussions, although it is sometimes phrased differently. It might be "How many NICs do you recommend for ESX?" or "How much RAM can you put in an R805?" or "Are blades the most power efficient server?". The really cool thing is that the answer to all of them ends up being the same - If you can tell me what you are trying to do, and the key requirements that you have, I can give you the best answer. Sometimes you need tons of RAM, sometimes you oodles of NICs, and sometimes you are looking for the best value.

We can give you some basic guidance and even provide you with reference architectures, but ultimately everybody has to customize their virtualization solution based on their needs. To provide you with the ability to be able to customize, you have to be educated about what your choices are what the ramifications are for each choice. That's what this focus topic on Selecting a Virtualiation Server has been all about. Connecting people with the information as well as each other for discussion about the process. This played out in each of the three chats that we did, and can continue on the discussion threads.

This really isn't anything new in the IT world. Almost all solutions have required a certain amount of planning and sizing that was specific for each organization. The advantage with virtualization is that there is more wiggle room to adapt and learn as you go. Because virtualization provides a layer between the VMs and the physical servers, things can be changed, modified, tuned much more easily than without virtualizaiton. If you find that a server is overloaded with too many VMs, you can use VMotion to move some to another server. If you find that you are out of capacity, you can add another server to the farm and redistribute the VMs to take advantage of the new capacity - without any downtime.

Use this community to keep the discussion going about what the best server is for virtualization - all questions around this topic are welcome. The opportunity to interact with your peers and find out why and how they made their decisions is really the best answer to a question that has a different answer for everybody.

Todd

Back in the ESX 2.5 days it was a common recommendation to have at least 3 NICs for ESX. One NIC each for the service console, VMotion, and VMs. It was really a waste for many to have to dedicate an entire NIC for the low network traffic to the service console. Then with ESX 3 the networking features became more flexible, and you could easily setup the service console to share a NIC with VMotion or the VMs.

The M600 and M605 blades for the Dell M1000e blades chassis now have two on-board NICs and the option to add two more I/O cards, each with two ports of either Ethernet or fibre channel. This means that you can have four NICs and two fibre channel ports for storage OR six NICs with 2 dedicated to iSCSI for storage. The four available NICs for VM data traffic is usually enough, although I admit there are always exceptions.

So if the I/O options on the blades are not enough there are options with the R805 and R900 and R905 that offer oodles of NICs. Specifically the R805 has four on-board NICs and four PCI slots. Filling these slots with your required mix of quad-port NICs and dual-port fibre channel HBAs would be a great solution for environments that have high requirements of physical NICs to support their VMware environment.

Todd