Why ESX are not suitable to run on Blade
Posted by craig 
- on September 30th, 2008 in Data Center, Hardware | 7 Comments »
1st of all, the reason of being utilize blade system in the market are looking at the point of servers consolidation, reduce power consumption and reduce the TCO require to purchase in term of hardware compare to the 1U, 2 U and 4 U servers. When we do compare the reason of having blade, you will always notice it was comparable between 2U and 1U servers in the x86 family and data center environment. In large scale deployment, you will always see that the Blade allow you to scale and spend in the sense with more stand alone machine you can have with the limitted rack space and power you do have in your DC. These seems to make sense for us to start moving to blade, BUT it also have some risk which will become major issue later on.
Before you can use blade, you require higher power consumption per rack to support approximately 30 to 32 blades per racks on 42 servers rack. At the same time, the cooling unit design in you DC require to be customize to ensure your blade chassis is working in perfect condition. Once you have this, then you may able to start think about Blade.
ESX on Blade have been some idea I personally thought before. The products I specify looking is the latest DELL Blade M1000e. The power and cooling in my DC is not a big issue. When I do analyzed the possibilities, I found couple of show stopper to deny my decision to move forward on that. As the enterprise architect point of view, Blade will be more suitable to consolidate those machine which require to run on physical VS Virtualization. The reason of that, is not really the matter of CPU or Memory you can have in the single blade, is really about the redundancy and performance we focusing on our virtualization. The limited number of pass through, NIC, and FC per blade is really not able to meet the number of VM we tried to achieve per host. We require redundancy, teaming as well as performance through put in term of networking and storage with the ESX servers we have. When we do calculate in term of cost per VM, the number had not show up as significant saving as we expected.
In additional to that, the more ESX machine we have in our Virtualization farm, it always require additional efforts to manage it for long term basis. There are many cases that users had built the ESX server with only 2 gigabits NIC for VMnetwork, which end up facing the performance issue in term on the Networking as well as the single point of failure. Virtualization is not performance sacrification. If we do plan properly, we will gain performance in virtualization vs under utilization physical machine.
Here is those finding I have and I will say that the Blade will not fit the virtualization requirement to achieve High availability and performance requirement. If we will have enough money to be spent on Blade environment, I believe you should have sufficient budget to go for something more suitable like R900, R905 and others 4U servers which provide more memory and CPU you need.
Tags: Blade, CPU, Data Ceter, dc, Dell, ESX, Memory, Servers, VM, VMware
7 Responses
So you have looked at one blade vendor and model and decided that blades as a whole do not stack up. Your analysis is not quantitative. What you are really taking about is scale-up vs scale-out. A good paper on this from IBM is http://www.redbooks.ibm.com/abstracts/redp3953.html?Open and the other vendors have similar ones.
Not saying I agree or disagree with your statement, just how you get there.
Thanks for sharing the white paper, the details it had mentioned had been part of my consideration in my experience. The limitation that we do concern is about the ratio you need to achieve example if we talk about 30:1 or 40:1, which large scale of virtualization. Bottleneck happen with the limitted number of network ports available. If you compare with the 4U servers which provide easily more than 10 Network ports to support the teaming, redundancy, HA/DRS, ISCSI and etc. We will not able to have 30 VM sharing for 2Gb or 3Gb or network connection as that will be major impact for performance within the VM.
Regarding the blade model, we had look into the HP blade as well. Many opinion are really how much performance do you look when you go for virtualization. We not looking to sacrifice performance to virtualize our environment, indeed, it does improve server performance when we virtualize it.
Again appreciate your opinion, is just depend what you want to be achieve. Today, I believe many environment, we no longer looking at ESX servers with 32GB as max, as the reason behind is common with quad core and 6 core is on the way. you will also see 8 cores on the way, with more CPU power you have, absolutely you looking forward to further utilize the computing power you have.
Did you take a look at HP’s new 495c virtualization blades?
http://h18026.www1.hp.com/products/servers/proliant-bl/c-class/495c-g5/specifications.html
they look like they would overcome some of the issues you mention, if you had 10GB Ethernet and 8GB fiber.. that could minimize some of the concerns.. but that is a pricey infrastructure. I find the Solid state drive an interesting option as well. If bandwidth is an issue.. you can also use LACP on the switch side to aggregate a larger number of uplinks into a larger pipe at least on the HP blades we have looked at.
In my opinion the suitability of blades also depends on your goals. In our case we are looking at the blades for ultra high density virtualization deployments, where we would deploy a large number of small servers, such as web servers, low utilization app servers and the like. If you were looking to virtualize high IO applications, then blades might not be your fit.
I think the biggest issue with the blades is the intial ramp up, once they are in they are quite simple and cost effective.. it’s getting over the hurdle of that first deployment, and intial financial outlay.
I would agree that blades aren’t the solution for everything, but then neither is virtualization. There are probably some environments that could benefit from vmware on blades.
Hi Jason,
Absolutely agree with the point you mentioned. Is again depend what is the best fit for every situation.
Look at this: http://www.dabcc.com/article.aspx?id=9114
The Cisco blades are looking pretty decent – at least from their specs. Can you say 384 GB of RAM?
http://www.datacenterknowledge.com/archives/2009/04/16/cisco-unveils-blade-server-memory-pricing/
this post was quite some times which referring to previous technology available on blade. As today, the blade environment should be good to go for virtualize as long as FCOE or HCA integration which provide virtual hba and virtual NIC available, and the memory per blade had been significant increase. But do remember to check out the detail of blade to ensure the chassis design is fully redundant to avoid whole chassis to be down at 1 time.