One of the types of advice I really appreciate is that which helps me to keep my options open.
I have a team from Dell in the office this week, configuring a giant pile of equipment we bought. The equipment includes a bunch of blade servers. We’ve relied on rack-mount equipment for decades, but with a push towards a private cloud we opted to jump into the early 21st century with blades. I’ve had relatively little experience with blades so it’s nice to have more experienced people around.
When I’m designing a system I always try to figure out what it’ll need to look like four years from now. Seeing the future is the hardest part of designing a system. Will our workloads increase? Will we need more CPU or storage? If we need more CPU or storage, how will we add it? Should we buy it all now anticipating the need or leave unused drive bays in an array to fill later? Will the drives be the same as the ones we have now? Will we have enough ports on the switches to grow? I even worry about things like whether the naming scheme will be flexible enough to handle growth. To compensate for this I try to give myself options, empty disk bays and RAM sockets, use 10 Gbps NICs instead of multiple 1 Gbps, etc. I’m pretty good at it with rackmount systems. Not so much with blades, yet.
Today I took our pile of Dell PowerEdge M620 blades and spread them between the M1000e blade chassis we have. The M620 is a half-height blade, so I stuck them all nicely & logically in the top row of slots, 1 through 8. I thought it looked great until a Dell guy, Tim, asked me if I’m ever going to put a full height or quarter height blade in the system.
My response, understanding immediately: “Oh, crap.”
The truth is that I don’t know if we’ll ever need a full height blade, like the Dell PowerEdge M830. Likewise with the quarter-height PowerEdge M420 and M430s, which sit four-tall inside a full-height “sleeve.” By filling the top row of the M1000e chassis I prevented myself from easily being able to add either kind. It’d be better to fill top & bottom evenly from left to right. And with a two minute change, suggested in passing by someone who has way more experience with blades than me, I kept my options open for the future.
Here’s a question for you folks: is it cheaper to make a bad or ignorant decision now and pay for it later when it needs to be fixed, or pay for the knowledgeable consultant up front so you don’t waste time and money later?
Stick your thoughts in the comments.
Comments on this entry are closed.
Actually, for HP chassis at least, you should fill them in top and bottom but also outer to inner slots for better cooling.
I’m surprised you go for Blade technology for your private Cloud project as nowadays people tend to go for rack-mount server for such project for the following reasons:
-Low entry cost than Blade especially if you don’t fill up right away your chassis
-Lower up-front cost (you need to buy a chassis with all the necessary interconnectivity modules before you can add a blade)
-More expansion capabilities
-Rack-mount server is commodity (Blade technology is not which will increase price/VM at then end)
-Training and support (You may have to train your team to support this new technology…)
-Smaller utility footprint (usually requires less cooling and power than blade)
-Standard powers supply (Blade chassis requires higher amperage and modified power plugs)
If you can get a knowledgeable consultant… use them. Mistakes made in the foundation of a platform lead to huge problems down the road. Note, I experienced cooling issues, make sure your monitoring is working for you might need to make floor tile adjustments.
It’s a tough call. On one hand we as Admins want to have as much information as possible before hand to be able to design systems the best/right way. Unfortunately, the reality is that the business/management can’t or doesn’t always provide us with that information and we are left to fend for ourselves and make the best possible decision, weather it be hardware or software. I have been fortunate enough to have gone through both situations, where I have had to make a decision and had to pay to fix it later and also have paid upfront for services. Both provided a wealth of knowledge for me. The first was a great learning experience in that I learned from my mistakes and learned as I was implementing the solution. No better way to learn if you ask me (sometimes). The latter provide me with a good understanding of the product I was working with upfront, so it left me being able to manage and administer it better, but I wasn’t the knowledge in installing and configuring it.
There is good an bad with each, I usually go with asking for forgiveness later. :)
We made the jump to blades about 3 years ago and only have 4 of our ESXi hosts left on rack-mounted systems. We have been happy with the move.
We chose the left-to-right layout as per your picture. The thing about set-up is you do it once and you rarely repeat it. We have chosen to develop a build script and configure it using the script for the next chassis.
I have always been intrigued by Dell blades, but we are a heavy “two-letter” shop.
I think we all try to avoid painting ourselves into corners as much as possible. But there will always be those decisions where, though you don’t know how or when, it will come back to haunt you at some point. It’s often hard to defend a hunch though, and so we stand, waiting for the inevitable.
It sounds like you’ve already got the bulk of your needs worked out. Where blades tend to get expensive is where you run the chassis past the end of sale date and you find you need a component (switch module, for example) that is no longer manufactured.
If you’ve got all of the parts you need for this generation of chassis, and you are satisfied with your on-hand spares versus service contract, you can’t go too wrong. Paying a modest amount to a knowledgeable consultant to review future plans and your current environment for any gaps is a nice way to get peace of mind. But you should be in pretty good shape – any situations in the future that require a different layout of blades in the chassis should be easily accomplished with vMotion (or your local flavor).