What server hardware to choose for your virtualization project?
Selecting physical server hardware while rounding up the business case is always a fun exercise, what do you need and what is smart too choose?
A question you should definitely ask yourself is are we going to scale up or to scale out or do both? With scaling up you will put more CPU power, Memory and network bandwidth in your physical servers to accommodate more virtual servers on it. With scaling out you just put another physical server box into the datacenter racks with the same dimensions as the other physical servers.
The pros and cons for scaling out versus scaling up.
Scale out
Pros for scaling out are that you just buy another box and spread the working load across one more box. You can have up to 32 physical servers in a VMware DRS cluster. The more servers there are in a VMware cluster the more efficient DRS will work. Also the overhead for High Availability, n+1 physical servers, is a pro for using more servers.
Cons for scaling out is that you will need to buy more licenses, buy more hardware support, use more rack space in the datacenter and you will have to go through the ordering a physical server workflow which will take time.
What server types will you use for scaling out, three of the most used servers are:
Dell PowerEdge III 2950
Dell PowerEdge R805
IBM x3650
IBM System x3655
HP ProLiant DL380 G5
HP ProLiant DL385 G5p
Scale up
Pros for scaling up are that the price per vServer will drop, you don’t have to use more datacenter space, hardware support costs won’t rise and if you don’t upgrade the server with CPU power license costs will stay the same. Also you have less hardware to manage in your datacenter.
Cons for scaling up are that if you run more and more virtual servers on a physical server and you want to perform maintenance and put the box in maintenance mode you don’t want to wait till 80 or more virtual servers are moved off. Same counts for when a physical server dies and High Availability kicks in you want those virtual servers back at work asap.
What server types will you use for scaling up, three of the most used servers are:
Dell PowerEdge R900
Dell PowerEdge R905
IBM x3850
IBM System x3755
HP ProLiant DL580 G5
HP ProLiant DL585 G5
Bottom line is to try and find the sweet spot where costs meet functionality, often I use servers with 2 Quad core CPU and 32GB memory (slowly moving to 64GB now prices for memory are dropping). Why not use 1U rack servers you ask? Ofcourse you can do that, but bear in mind that you have to fit NICs and HBAs in the little box. With the standard setup we often chosse 1U high servers just don’t have enough space to hold 2 or 3 NICs with quad ports on them and also fit 2 HBA’s in them if you use a Fiber Channel SAN.
You are wondering where the blade servers are in this comparison, I will blog about the choice between normal rack servers versus blade servers in the upcoming days.
Edit: I have added the AMD server types to the links.
Update: Links no longer active
8 Comments
Leave a Reply
You must be logged in to post a comment.
So what about AMD vs Intel. And why not pick the Dell PE R805 over the R900? You could have a higher rackdensity and a lower TCO?
So what about AMD vs Intel. And why not pick the Dell PE R805 over the R900? You could have a higher rackdensity and a lower TCO?
Following on from Duncan’s comment what about the DL385 G5 which has more memory slots and expansion options than the DL380. I know the examples are there to be like for like per vendor, but the DL385 is a step above the DL380 currently in terms of using them for virtual hosts.
Cheers
Following on from Duncan’s comment what about the DL385 G5 which has more memory slots and expansion options than the DL380. I know the examples are there to be like for like per vendor, but the DL385 is a step above the DL380 currently in terms of using them for virtual hosts.
Cheers
Duncan and David, you both got some good points there. There are a lot more models to choose from including AMD-V technology or Intel vPro technology. Wasn’t my intention to just provide only Intel models, because AMD models also rock like the HP DL385 or HP DL585 for example. Also the Dell 2950 will be replaced very very soon.
What I would like to get out is choose a model which gives you all options you need for the best buck, but look at them with the scale out or scale up in your mind. We have placed HP DL585 machines with customers, were they would have been helped with HP DL385 machines.
Also would be nice to make a post about the whole AMD vs Intel around virtualization ;-)
Duncan and David, you both got some good points there. There are a lot more models to choose from including AMD-V technology or Intel vPro technology. Wasn’t my intention to just provide only Intel models, because AMD models also rock like the HP DL385 or HP DL585 for example. Also the Dell 2950 will be replaced very very soon.
What I would like to get out is choose a model which gives you all options you need for the best buck, but look at them with the scale out or scale up in your mind. We have placed HP DL585 machines with customers, were they would have been helped with HP DL385 machines.
Also would be nice to make a post about the whole AMD vs Intel around virtualization ;-)
In my opinion, there are some other issues you need to address while scaling up and scaling out.
First of all, when scaling out, you need to keep a close eye on the compatibility. One server isn’t the other, as we all know. Usually, when upscaling or outscaling is in the picture, the farm is not quite up to date. At the rate the manufacturers release new architectures, the risk of incompatibility is most certainly an issue.
Also, when scaling up, you need to ask yourself if it’s really useful. If a server is nearly 2 years old, it might be a disinvestment to add additional hardware to it, keeping the lifecycle in mind. Most manufacturers do not lower the prices for additional hardware so a CPU mostly costs as much at day 1 as it does at day 721 in the life of a server.
So, all in all, scaling right is the real magic, if you ask me. I think upscaling or outscaling is only useful in a handful of cases.
Greetz,
Froos
In my opinion, there are some other issues you need to address while scaling up and scaling out.
First of all, when scaling out, you need to keep a close eye on the compatibility. One server isn’t the other, as we all know. Usually, when upscaling or outscaling is in the picture, the farm is not quite up to date. At the rate the manufacturers release new architectures, the risk of incompatibility is most certainly an issue.
Also, when scaling up, you need to ask yourself if it’s really useful. If a server is nearly 2 years old, it might be a disinvestment to add additional hardware to it, keeping the lifecycle in mind. Most manufacturers do not lower the prices for additional hardware so a CPU mostly costs as much at day 1 as it does at day 721 in the life of a server.
So, all in all, scaling right is the real magic, if you ask me. I think upscaling or outscaling is only useful in a handful of cases.
Greetz,
Froos