Customers clearly want the kind of flexible IT offered by cloud computing. But they don’t want to lose control over their resources and service levels. Moreover, IT vendors have traditional revenue streams to protect.
Enter the Windows Azure Platform Appliance. Customers can try Microsoft’s cloud platform on hardware they physically control. System vendors can sell their equipment as modular cloud “building blocks.” Microsoft can continue selling its software through well-established channels.
So what does this mean to cloud rivals Google and Amazon? In terms of public-cloud market share, not much. After all, the new appliance is clearly aimed at private- or hybrid-cloud infrastructures.
But cloud providers that dismiss Microsoft as a threat do so at their peril. The Windows Azure appliance affords Microsoft the kind of foot in the door it has used many times before—very effectively.
I recently had one of those “everything you know is wrong” moments. I’ve used VMware Workstation on Windows since version 3.0. One performance rule of thumb has remained constant through the years: “Avoid disk I/O at all costs.” And, of course, the corollary “Avoid Guest OS paging I/O” has been a key part of obeying the first rule.
That translated into a balancing act for me as a user. I needed to give the guest VM as much memory as possible – but not so much that the host OS start paging excessively. The balancing act got harder when I was running applications on both guest and host. When it got too hard, the easiest answer was usually “buy more memory.”
I recently purchased the latest version of Apple’s Macbook Air, the 2.13 GHz model with a 128 GB solid-state disk. First off, I need to say that I’ll never go back to a magnetic HDD laptop. The improvement in the machine’s responsiveness is that dramatic.
But here’s the thing: a Macbook Air comes with 2GB of RAM soldered onto the motherboard. Ok, that mitigates the risk of a cold boot attack, which is good for someone like me that travels a lot. But I need to run apps on Windows, especially Outlook. That meant I was back in the guest-vs-host memory balancing game.
For a variety of reasons, I run Windows 7 in my guest VM. Windows 7 has a published minimum RAM requirement of 1 GB. The Air doesn’t really have a full 2GB available for me to play with; some memory is used by the onboard video chipset, for example. And, of course, VMware adds some of its own memory overhead. I need to run apps both natively in Mac OS X and in my Windows 7 Fusion guest.
The balancing act got hard. “Buy more memory” wasn’t an option.
So I started nibbling at the edges. I discovered I could disable hard disk I/O buffering in VMware Fusion’s advanced settings. I got some host memory back, and didn’t suffer any appreciable performance loss. Then I nudged down the guest RAM to 896 MB. No significant performance difference.
Meanwhile, I’d migrated most of my work onto Mac native apps, and was primarily using Windows to run Outlook. Some performance monitoring within Windows showed I was typically using only 512 MB. What if I were to do something radical, like drop the memory allocated to the guest down to only 512 MB?
Documentation published by VMware – and just about everyone in the blogosphere – says that 1 GB really is the minimum needed to get reasonable performance from that guest OS. Dropping the RAM allocated to Windows 7 on VMware Fusion would be a performance disaster.
Not on an SSD based Macbook Air.
An SSD lets us turn the rule on its head. The horrendous I/O bottleneck inherent in laptop HDDs has been removed. My rule now is: “Let the guest page.” Really. It’s OK.
Has anyone else tried this?
Recent Comments