xen

The last few weeks have been very exciting for the virtualization market. Here are some highlights:

First, the VMware IPO created a lot of waves: amidst skyrocketing stock prices and huge oversubscription, this IPO was right up there with the Google IPO in the hype meter. It is being called the year's best IPO for a good reason – in relative terms, this IPO was indeed huge, closing at 76% higher than the initial prices at the end of the first day.

The repercussions of all this money influx can already be seen. Silicon valley has been haunted the last few years by Google's hiring spree: for a long time it seemed like the default destination of anyone smart in the bay area looking for a job was Google. They hired good people, and they paid well. It was scaring a lot of the big players in the market. Well, now it seems that VMware has joined the club, with jobs paying between 130-160K. Phew!

A day after VMware's IPO, software giant Citrix announced its acquisition of XenSource (see press release). Needless to say, the timing of these events was unlikely to be completely co-incidental. Surprisingly, the Xen community has been awfully quiet about the deal – the development mailing lists barely touched upon the subject. I'm not sure if thats good or bad, only time will tell.

Meanwhile, in a series of recent posts, the GNU libc maintainer, Ulrich Drepper has criticized the hypervisor approach taken by both Xen and VMware, and instead thinks the KVM approach is a much feasible route to efficient virtualization. More on this later.

and quite expectedly, it outperforms Xen. I just finished reading another white paper last night which bashes Xen. I have more to say on that white paper, but the short story is that I think the paper is extremely biased and tries to convince the reader that VMware's products are the best, but in the process also spreads a reasonable amount of FUD.

Anyways, coming back to the aforementioned benchmark – what I find most ironic is that VMware has such strict policies on third parties conducting independent benchmarks, and here they are tooting their own horn. I'm sure ESX performs great, but wouldn't it be more believable if researchers could independently verify this claim? Until sometime back this was near impossible. Though it seems that recently VMware relaxed its licensing restrictions a little bit, so that internal benchmarks can be made public once the assumptions, methodology and results have been verified/approved by VMware.

A few weeks ago, Amazon quietly launched the Elastic Compute Cloud service (beta, of course). I had been dying to get my hands on an account since then and just today I got my confirmation email. I wasn't really expecting something like this from Amazon (though after S3, it wasn't too unexpected), so I was even more intrigued. And when I found out that EC2 was running off of Xen, I was just blown away!

Anyways, so here are my notes so far:

• the documentation is excellent, so I won't waste your time and mine going over it again. Everything worked out of the box for me.
• Yes, they are running Xen.
• The getting started image comes with 9G of disk on the root partition and an additional whooping 140G on /mnt. /mnt doesn't seem to be network mounted (unless its a networked block device) though.
• The VM has a lot of RAM. My VM shows me 1.8G of RAM. Thats more than what I have on my laptop!!!
• The VM sees a 2.4-GHz uni-processor AMD Opteron. So Amazon is not using Intel machines for their cluster. Thats interesting.
• As far as I can tell, they are running just two VMs per physical machine, with 50% hard reservations for each of the VMs.

More later.