Virtualization.info recently posted an link to an article on InfoWorld stating that Q3 server sales are down and that Gartner attributes this to virtualization. I think that this trend is still in the infancy stage. I work with a lot of companies who are still just starting to play with virtualization.

Over the next few years I think that virtualization will become a standard in the industry. As a result this will impact companies like HP, IBM, and Dell resulting in fewer total server sales, however I this doesn’t have to mean the end of these companies. As more and more companies adopt virtualization as a standard, they will stop needing lots of small server. Instead they will move away from 1 & 2 CPU systems to 4 & 8 CPU systems.

It will be interesting to see how Dell will react to this. Dell has a reputation for being extremely aggressive in the server sales arena, many times taking a loose on margin just to win a deal.

Off the topic: Don’t think my silence about the recent Microsoft / Novell agreements means I don’t have an opinion about it. My head is still spinning and I need a little time to develop an appropriate statement. I will say however, that I think this may bring about the first “fork” in the Linux source.

I’m embarrassed to admit that it’s been 11 days since my last post. Between some shake-ups on the job front and the Thanksgiving holiday the time just went by too quick. However, things are back on track and the posts can start flowing again.

On a funny note, I tried to get support for an application I use on my Treo 650. I needed help with a problem using Goodlink, so I went to the Goodlink website and located their phone number. When I dialed it, I received an automated recording that went started something like this:

“We are in the processes of moving our office and will be closed from June 27th to July 1st…”

It then proceeded to hang up on me. I couldn’t help but laugh. How many calls have they lost over the last 5 months because nobody noticed that the auto attendant is out of date? Too funny.

I just found this on YouTube. This really sums up it up.

I wasn’t able to attend any sessions today, I had to drive some friends to the airport. I just wanted to take a momend to thank everyone at VMware for a great event. VMworld 2006 was very fun, and there were a lot of extremely educational break out sessions.

I’d also like to thank the many companies who participated in the Solutions Exchange. It amazing to see some of the solutions people have come up with to better leverage Virtual Infrastructures.

Tags: VMworld 2006

That last post about the buses was from my phone. It wasn’t my best post ever, but I was a little annoyed. However, after being there for a few minutes, all was forgiven. I really had a great time.

The lines to get in the rides were short and the rides themselves were fun. My personaly favorite was the Mummy, with Back to the Future a close second.

The other nice thing about the event was the change to kick back and shoot the breeze with some of the folks from VMware. I live in California, but deal with folks from across the country and don’t always get face to face interaction. This event was a good chance to hang out a little and get to know people.

VMware: Great Event!

Tags: VMworld 2006

OK, this is stinks. Tonight’s event is supposed to be at Universal Studios. The buses were supposed to pick people up at their hotels at 6:30 PM.

I’ve been out here since then and there have been no buses until 7:05 PM, and it was FULL!

Luckily there was another bus right behind it.

I know it’s hard to coordinate transportation for 7000 people but this is nuts.

Session title: “Intel and VMware: Accelerating & Simplifying the Move to Virtualization with High Performance, Low Power, Dual-core and Quad-core Computing Platforms Utilizing the Technical Best Practices and Reference Configurations from Virtualize ASAP”

Although the title was extremely bloated, this session wasn’t. It was very clear and informative regarding the benefits of multi-core processors and how newer virtualization features can benefit users.

The session started with some industry statistics of which I found the 2 most interesting to be:

“79% of IT budget spending is to keep the business running” 79% seems like an outrageous number, so I’m assuming that this includes things like kWh, HVAC, etc…

“$3000/Server 4 year electric cost with 20% annual growth”

After laying the groundwork and exponding on Intel’s processors we got to the meat of the presentation surounding Virtualization enhancements.

Although the presentator didn’t go into deep technical detail, he did explain how EPT (Extended Page Tables) will allow the VMM to freely modify the underlying page tables thereby providing optimized access to the Guest OS.

The other virtualization technology that was highlighted was I/O virtualization. This “defines an architecture for DMA remapping”. This is going to give the Guest OS direct access to physical hardware device.

I would have liked more detail on these new technologies…

Tags: Intel, Virtualization, VMworld 2006

Session title: “VMmark: A Scalable Benchmark for Virtualized Systems”

When I heard about VMmark and that there would be a presentation on it here at VMworld I was very excited. For many of us, proving the performance capabilities of Virtual Machines within ESX has been a challenge. We’ve all come up with our own methods of showing how powerful a Virtual Machine can be. With VMmark, this process is going to be much easier and more reliable.

VMmark is intended to:

Measure the perfornance of hardware platforms

Be run by hardware vendors and virtualization software vendors

Help study the effect of architectural features

VMmark is not meant for:

Capacity Planning

Specific non-standard cases

VMmark consists of “tiles”. Each “tile” consists of 6 VM’s running different operating systems and workloads to simulate a complex and diverse environment. Those systems are:

A Windows 2003 file server

A Windows 2003 mail server

A “standy by” server running RedHat Enterprise Linux 3 (RHEL 3)

An Apache web server running RHEL 3

An Oracle OLTP database running RHEL 3

A java order entry running on Windows 2003

It’s nice to see this blend of VM’s. This helps show a more realistic represendation of what a real Virtual Infrastructure might look like.

I’m looking forward to being able to work with this.

Tags: VMmark, VMworld 2006

Session Title: “Citrixand VMware: How these Two Technologies Work Together to Provide More Efficient Computing”

So far, this has been the best breakout session I’ve attended. The presenters did a lot of extensive testing of Citrix Presentation Server inside a VM on both ESX 2.5.X and 3.0.X.

The Physical server that they used was a HP DL585 with 4 AMD dual-core Processors and 32GB of RAM. The VM they used was had 1vCPU and 3.5GB of RAM, running Windows 2003 (32bit), along with Citrix Presentation Server 4.

Their test script did the following loop:

Opened a MS Word document and typed for 11-15 minutes.
Paused shortly then repeated.

This script simulated a single users and a new instance was started every 30 seconds. This helped to simluate multiple users accessing the Citrix Server.

This test was done on both ESX 2.5.X and 3.0.X. The results were interesting:

With ESX 2.5.X 80 users ran the CPU at 85%

With ESX 3.0.X 140 users ran the CPU at 80%

(Great Kudo’s for ESX 3.0.X)

The presenters were very clear that they weren’t trying to say you can get 140 users on a Citrix VM, individual results may vary depending on what the users are doing. It also shows the performance improvement from 2.5.X to 3.0.X.

To get optimal performance from a Citrix VM the presenters offered the following tips:

* In 2.5.X the Terminal Server Switch was needed, but not in 3.0.X

* You should disable Transparent Page Sharing. Go into Advanced Setting and change the following:
Mem.ShareScanTotal = 0
Men.ShareScanVM = 0

Note: This will effect the entire ESX server

* Change:
Mem.ShareScanThreshold = 4096

(Sorry, the slide went to fast to write down their notes on why.)

* Dual Proc VM’s are not recommended. Multiple vCPU’s create more of an overhead than it’s worth.

* Disable Hyperthreading, although they won’t guarantee that this will improve performance. You can either disable hyperthreading for the entire ESX server or for a specific VM by editing the vmx file:

shed.cpu.htsharing = none

* Use the LSI Logic SCSI adapter.

* Disable COM, LPT, & USB

* Disable auto-detect of CDrom

* Disable Visual Effects in Windows

* Do not over allocate RAM. If the server has 32GB of RAM, only allocate 32GB for the VM’s.

Session title: “Using VMware and Neverfail to Implement Cost Effective Long Distance Application Failover”

Neverfail as a software didn’t really excite me. I had heard some buzz around it lately, but the presentation didn’t do anything for me. As a failover clustering software it looked OK, and if it works as good as they promise then I can see value for it. However I was disappointed that it only worked with Windows. This is a big limit for me.

In the perfect world, I would like a clustering application that plugged in to ESX and replicated VM’s regardless of their Guest OS (like esxReplicator). It could just capture the I/O and replicate it to a warm failover site. Then actually do the failover in the even of a site failure.

As for the session there were a good quote I took from it:

“Companies lose up to 16% of their annual revenue due to downtime.”

Tags: VMworld 2006