Cisco readies move into the server hardware market

I’ve bought “Cisco” servers many times before.  They were always HP or IBM servers with a Cisco logo tagged on the bezel and came with ridiculously high support contract and spare part costs, but they were Cisco servers.  It seems that Cisco was not satisfied with playing at that level, however, and now they are moving forward with a true line of Cisco server hardware, part of their Unified Computing System platform.

The newest component of this platform – the servers – is a blade server chassis designed to plug in to the existing fiber and copper switching infrastructure as well as work with their storage management tools.  It all sounds great, I suppose, but I’m not entirely sure that the move will be compelling for an enterprise.

For starters, the hardware platform is all unproven.  Sure, the vast majority of the parts in any of these systems are OEM from known manufacturers but there is still a ton that goes in to the final assembly of those systems.  After all, how else can you explain just how badly the Dell Blade systems are assembled relative to the HP or IBM blades?  And even if they have better and cheaper hardware (like Sun did with their x86 server platform) , there is still a huge unknown about how well Cisco will be able to support the server infrastructure.  They are generally very reliable when it comes to supporting their switching hardware, but after that things tend to tail off. 

And, of course, since this is the era of virtualization, it is critical that any story about data center infrastructure include mention of the ability for the platform to be “fully virtualized” or support non-virtual solutions.  I love marketing buzz.  The hardware will be available with both VMware and Microsoft Hyper-V platforms, so they are continuing to partner with the big players on that front.

Only time will tell just how astute a move this is from Cisco, but there are certainly plenty of open questions (like when they are actually going to have hardware available in the channel) to ponder in the meantime.

Virtualizing the I/O layer

Hardware and software virtualization are old hat at this point, thanks to VMware, Hyper-V, Softricity/SoftGrid, Altiris and many others in those spaces. So, lacking anything else new in those areas, some vendors have decided to hype the virtualization of the I/O layer, focusing on the "virutalization" of ethernet and SAN connectivity. I found an article/post about the topic from last week that talks about a couple vendors, 3Leaf and Xsigo, but neither the article nor the vendors are particularly up front on what they are actually doing. Their products are InfiniBand switches, plain and simple.

Don't get me wrong - I love the InfiniBand concept. But seeing these new vendors popping up with their "new" technologies is rather silly in many ways. InfiniBand basically extends the PCI plane from the server to a switch via a 10 Gbit connection. In the switch you can have multiple connections out to various other services, most commonly Ethernet NICs and Fiber Channel SAN connections. InfiniBand is getting more interesting now that the wiring and standards for the 10 Gbit connections are pretty easy and common, and also since 4 Gbit SAN connections are available, but it definitely isn't new technology. Cisco has been selling InfiniBand switches for at least 3 years, and the protocol goes back older than that.

Of course, the fact that the technology has been around a while actually is a good thing when looking to evaluate it, either in the form of these new vendors or from an established player like Cisco. Instead of multiple switches for each topology, everything runs from the server to the switching infrastructure on a single InfiniBand connection (two, for redundancy, if you're into that). Instead of installing various drivers to support all the hardware there is a single driver for the InfiniBand card. And from the InfiniBand switch you can provision multiple NICs and multiple HBAs on a server, all through the one InfiniBand connection. It will definitely reduce cabling complexities and provisioning time, so those are both good things.

Definitely an interesting technology to take a look at again, but also most definitely not new.

Dell (finally) upgrades their blade offering

After lagging behind HP and IBM for several years, Dell has finally made a move in the blade server space that appears to make them actually competitive. The new PowerEdge M1000E enclosure and the PE M600 and M605 blades are comparable to the c-series from HP and IBM's BladeCenter H. They include 16 blades in a 10U chassis (comparable to the others), and similar spec's on total RAM, processor options and storage as the base models from HP and IBM, so there is some reasonable performance available. That being said, there are no options for the "double-wide" blades that HP and IBM offer, allowing for more processing power in a single computer (quad processor, extra local disk, etc.).

It is a 200v+ power configuration only, so not viable for small offices/remote locations (HP has a c3000 chassis that will run on 110V power), but it has the full complement of Ethernet switches (Cisco and Dell), redundant management cards, local KVM and Fiber Channel options available.

Dell still has some ground to cover to be completely competitive, but this is a great move on their part. Now they can't be dismissed without consideration because their product line is much closer to meeting the needs of many customers.