This week I had the unique opportunity to spend some quality time with two of my favorite people: Tony Redmond and Brian Desmond. The occasion was that we were invited by Hewlett-Packard and Microsoft to take an advance look at their new E5000 messaging system before its formal unveiling. Our goal was to spend some time with the program managers on the E5000 team learning about the system and experimenting with it a bit. The kicker: the whole process was videotaped.
Of course, nothing is as simple as it first seems. Tony, Brian, and I thought that “videotaped” meant “hand-held camcorders.” We were in for a rather rude surprise when we were greeted by Jenny, our makeup artist:
She wasn’t rude in the least, but boy, were we surprised to see her. Jenny quickly got our makeup on and onto the set, where we found no fewer than four large, dolly-mounted video cameras, each with a camera operator, plus a sound guy, a director, and assorted other studio folks. We very quickly got used to their presence and sat down with Dean Steadman and Karl Robinson of Hewlett-Packard and Jeff Mealiffe of Microsoft to talk about the specifics of the E5000 series.
Tony’s writeup gives a good summary of our discussions. Going in, I was of the opinion that having a solution that combines two DAG nodes in a single physical enclosure was useful, but that it was lacking in three key areas: it still needs load balancing; it introduces the potential for the E5000 to be a single point of failure, and it doesn’t include Exchange server licenses so it isn’t really a complete solution.
However, once I had a chance to talk things over with Dean, Jeff, and Karl, I started to come around.
First, load balancing. Yes, you still need it. In this respect the E5000 is no different than any other set of Exchange 2010 servers. I’d love to see load balancing integrated into a future revision of the E5000 series, as I think the customers who will be most interested in the idea of an all-in-one appliance are least likely to want to deal with the complexities of CAS load balancing.
Second, HA. It’s true that if you put an E5000 in your server rack that a site-level failure (including mains power, fire, flood, and so on) can kill both nodes at once– but this is no different than the situation faced by the majority of Exchange 2003 sites now, the very customers at which the E5000 is targeted. Replacing a set of Exchange 2003 servers with a set of E5000s means that customers can not only take advantage of Exchange 2010’s I/O improvements to reduce the total number of servers, but that they can also use DAGs to reduce the risk of data loss from a disk, controller, or server failure. Where business reasons require protection against site-level failure, it’s a simple matter to drop one or more additional E5000s in a remote site and join them to the existing DAG. The more I think about it, the more I like the idea of using building block-style hardware like E5000s as the basis of building remote sites. There’s quite a benefit to standardization of hardware, software, and operations.
Third, license bundling. Because the E5000 will primarily be sold through HP’s channel partners, I’d expect that customers who need Exchange server licenses or CALs will be able to order them at the same time as their E5000 units. The E5000 Quick Deployment tool won’t proceed with installation until you enter a valid Exchange PID. This is required because some models of the E5000 line will be configured with more than 5 databases, and to prevent unpleasantness the installer wants to verify that the PID is for the Enterprise Edition of Exchange. Making things a bit more complicated, from a business standpoint, Exchange server licenses have never been previously bundled in the way that Windows Server licenses have. I’m hopeful that we’ll see the option to bundle all the needed licenses in with the product. (In fact, I had a great discussion with the Microsoft folks about the desirability of an in-app purchasing system for CALs– imagine being able to buy additional CALs or server licenses as easily as you buy songs on iTunes or points on Xbox LIVE.)
The deployment experience is quite smooth; all of the questions normally spread throughout the Exchange setup process are consolidated onto a single page. Once you answer them, the setup tool does the rest. Bear in mind that you have to complete installation on the first blade before you can do anything with the second blade. Don’t give in to the temptation of starting setup on blade 2 before it’s done on blade 1. Trust me on this.
However, if you want to do things like pull out disk drives or even blades while the E5000 is running, go right ahead. I got to experiment with doing exactly this and it was quite a lot of fun. Bear in mind that there are two hot spares allocated for each blade, so if you pull a spare nothing will happen. However, pulling a data drive triggers a rebuild of the mirror pair containing the affected DAG, and then pulling another drive causes a DAG failover, just as you’d expect.
Tony, Brian, and I left the HP folks with quite a bit of feedback on every aspect of the E5000, but that’s why they asked us to come. The basic product is quite solid, and I expect it to be warmly received in its target market of organizations with 500-3000 mailboxes. It’s definitely a major upgrade over what customers in that segment have with Exchange 2003. On to March 1st, the official release date!