The Blog is Back!and I Made a Change!
Learn more featured post
October, 2009

Blade Servers or Rack Servers?

I’m still fighting off a cold after like 11 days, so I haven’t been writing too much.  Luckily I think I’ve finally got my voice back so maybe this week will be easier than last.  Some of those 3 hour meetings and presentations were rough!

My favorite type of engagement we do at Varrow is the Plan & Design.  It’s an engagement where I get to work very closely with the customer and dig in to their environment.  We do them for all sorts of technology implementations but usually it’s VMware vSphere and/or Site Recovery Manager.  It’s a way to really dig in to their whole environment, servers, platforms, storage, networking, business processes, applications, etc.  Many times during this engagement they’ll ask us for a recommendation on platform for VMware, and often it’s blade or rack?

Like a good little technical consultant my answer is “it depends”.  There really is no right answer to this question.  It’s really up to the customer and how they want to manage their infrastructure and server platforms.  Some people love blades and some don’t.  Blades can give you very good density.  While they often can’t have the amount of memory as a rack server they usually atleast go to 64GB which is the maximum for most people’s vSphere servers.  They allow for very fast deployment by just sliding in new blades whenever needed.  Also, they are often more power efficient since all blades utilize the same power supplies and cooling fans.  You gain efficiencies of scale.

The problems come in when you look at deployment costs and management.  That first blade costs you a lot of money since you have to buy the chassis as well.  I’ve found that the most successful method for deploying blades is to look at the chassis as “layer 1” type infrastructure, like you do with 42U racks and cabling.  This is especially true when you look at organizations that have different departments buying their own blades.  No department wants to be the one that has to buy that chassis so IT has to find a way to provide that as basic infrastructure. Management can also be an issue.  In a blade system you are adding another layer of complexity and configuration, especially around the areas of I/O.  Some people see this as a benefit and some see it as a hindrance.

Finally, you have reliability.  It’s rare, but entire blade chassis do fail and if that happens you just lost a lot of hardware at once.  This can also complicate your VMware design to make sure that things like HA and FT are configured properly so that losing a chassis doesn’t take down the redundant systems as well, even if you have multiple blade chassis.

In contrast rack mount servers are simple.  One server to one chassis.  While not as efficient for power and cooling, and usually physical space, you maintain that autonomous environment.  The I/O configuration is more straight forward and normally you can get more hardware in a rack server than on a blade.

In my opinion the benefit of conventional blades are not worth the extra layers of complexity, management, and cost.  Now, there are things changing that might sway my opinion, such as Cisco UCS.  UCS provides for very good I/O throughput and server hardware on the blade.  Pricing on them is also in-line with rack servers plus you get a very good unified management system to manage your entire UCS environment.  Time will tell if the other hardware vendors follow.

1 thought on “Blade Servers or Rack Servers?”

  1. Rack servers provide more expansion slots for I/O and are a perfect fit if internal storage is required for the servers. For instance, if the requirement is to run Virtual Machines (VMs) using internal storage, rack servers are a better choice because they have additional storage slots for internal storage. Also, if there is a requirement for a higher compute capacity per server, rack servers are the preferred choice as they can support 8 or more sockets. This works well if there is a need to run a smaller number of high power servers for virtualized workloads.

Leave a Reply

Your email address will not be published. Required fields are marked *