Architecture for Blade Servers.

[JagannathS]

I have 4 blade servers. I am planning to deploy esxi. My design is as follows. 1 blade for management and other 3 for VM. The 3 blades will form a cluster. In management i am planning to deploy vCenter for managing the other 3 blades. Is the design right? What are the components to be chosen along with vCenter in other words which suite from VMWare should i choose?. ( I need vSAN, DRS, HA, SRM features)

[Abhilash Basavarajaiah]

Is this for production or for lab?

That usually is right but what about redundancy if the management server fails? We segregate resources into management and resource clusters when we usually have cloud environment. When it's a virtualization environment, the management/infra servers would be AD, DB, DHCP, DNS etc with your vCenter and other VMware related products like vShield, SRM etc.

If its just vCenter that you are using and other infra servers then it's better to have them in single cluster so they can failover to other hosts.

So to not confuse you and put it in simple words, your architecture is fine if its production but you should also think about management host failure as its single point of failure if all infra sit on a single host.

Most of the features like HA, DRS, VSAN come with vCenter and ESXi itself. For SRM you would need a seperate windows machine.

If its lab then put all of them in single cluster so they can fail over during host failures.

[JagannathS]

Thanks Abhilash,

It is for Production only.

I am planning to use Chained mode for vCenters. So if the local vCenter Fails the remote vCenter can be used. Can this be done?

[Alistar]

To add to Abhil's response, be aware that you will have a single point of failure if you are using the blades - and that is a blade chassis. Although these faults are rare, we had it once in our environment and it took all the blades down and we had to wait until the technician arrived and replaced it, which meant about 4 hours of downtime. If you can spread the blades between 2 enclosures evenly, that would be for the best, but I doubt you'd have that luxury - perhaps in the future when the enclosure is being close to full.

Also if one host in the cluster would sit there only for management, it would be pretty much under-provisioned. Should it fail, you would lose all your vCenter capabilities. I see no problem with running it inside a cluster of all 4 hosts, using the vSphere's HA feature.

For your requested editions: To use DRS you need to have Enterprise Edition at minimum, and Enterprise Plus if you want to use Storage DRS. Keep in mind that SRM and vSAN are licensed differently. Take a look here for vSAN licensing and here for SRM licensing.

If you could share some hardware details of your blades/storage we can spark further discussion :smileyhappy:

[JagannathS]

Yes i accept the fact of a single point of failure. But in using chassis with 4 blades, i get the advantage of dual switch blades which gives me various other advantages.

So u suggest instead of dividing it to 2 clusters i put all blades in one clusted? in this case if the host which contains the vCenter fails will it move to other host and restart by itself? I am planning for Enterprise Plus edition.

What details of the hardware do you need me to share?

[firestartah]

Seeing as you only have four hosts I would recommend one cluster as if you only have one host for management then resources on that host may be constrained or wasted whilst the "consumer" workloads are all running on the other cluster.

Make one cluster but ensure restart priorities are set for the management components, this way you can benefit from HA and DRS and still get maximum usage of your hosts. Once you have more blades then separate into management and resources with at least three hosts in management (N+1 HA in a two host cluster is pointless)

Gregg

[JagannathS]

thanks Gregg.

Planning to make use of vSAN capabilities. Will you suggest the same even with vSAN?

[Alistar]

I absolutely agree that blade infrastructure will provide you with good network redundancy, just to keep the blade enclosure maintenance in mind as well (Onboard Administrator ugprades etc.).

Yes exactly, just create 1 cluster - in any case a host fails, the High Availability agent present in VMware Standard Edition will restart all the failed VMs on other hosts. Are there any Enterprise Plus edition features you will need to use in your environment by functional requirements? It might be an overkill for a "small cluster" - however this is completely another story if you plan to expand rapidly. Just thought you can save money there and put them for beefier hardware for example.

Type of the blade, memory and CPUs used along with NICs (speed & amount of ports you have available) would be neat :smileyhappy: Also what storage do you plan on using?

[JagannathS]

Yes Would require Distributed Switch, Host Profile & auto deploy, Flash Read Cache/ write buffer are required which are available in Enterprise Plus editions.

Configuration are as follows:

All blades Are Dual 8 core xeons with 96GB Mem per blade

Storage i am planning is 2 TB on the RTM wth 128 GB SSD for read Cache/ Write buffer.

Networking: on board 2x 1GigE, and 2x 1GigE to the Switch Blades, 2x 40GigE to Switch Blades.

Is there anything else u need?

[Alistar]

Okay great, thank you for sharing the info.

Can you please explain "2TB on the RTM" bit? I am not familiar with this - will it be localy attached storage or a SAN/NAS?. Also, please note that according the amount of the workload/VMs you have quite low core:memory ratio (that is if you won't be running CPU-intensive applications all the time).

Your networking is pretty fine, I suggest using the 1GigE for your Management network, and 40GigE for vMotion, Storage and VM traffic. You have some pretty nice options for configuring NIC failovers.

So I'd say you are all set for installation and clustering of your blades :smileyhappy:

[JagannathS]

Oh Sorry.

I am using ATCA Blade Servers. So there are Two components 1) a main Server (this is the Compute) and a 2)Rear Transition Module(RTM) ( this is usually for I/O Extension or Storage). It is a local Storage for a blade server which sits on the RTM.

Ya even my networking plan is the same as you suggested.

I could not comprehend the below statement. Can you please explain? Also, please note that according the amount of the workload/VMs you have quite low core:memory ratio (that is if you won't be running CPU-intensive applications all the time).

Also if you have any deployment guide of VMWare on blade servers can you share with me?

[Alistar]

Thank you for an explanation, handy to know another architecture.

Well, you have 16 cores per blade but only 96GB of RAM. Depending on your workload you might hit the RAM limit much sooner than an ideal CPU usage - the processor would be pretty much under-utilized. You could run many more VMs on one ESXi host with two or three times the memory. Of course if you need to run CPU-intensive workloads that don't require that much RAM, this configuration is completely valid.

There is no "right way" of deploying on blade servers - that is the charm of the whole VMware solution when you craft your own design according to the application's / businesses' needs. I am currently reading an awesome book which can give you more insight if you want to spend some money on it: http://www.amazon.co.uk/VMware-vSphere-Design-Forbes-Guthrie/dp/1118407911/

[JagannathS]

No according to application requirements 240 GB of memory will be sufficient. I have kept 50% extra RAM in each to help for the cause.

Thanks for the suggestion of the book. will definietely look into it.

Also can i make use of vSAN capabilities in the architecture?

[Alistar]

Ah all right, thank you for the clarification, you can just start setting up your environment whenever you are ready :smileyhappy:

I think you can, unfortunately I haven't got any experience with vSAN but please make sure your storage device is on its HCL VMware Compatibility Guide: vsan and a quickstart guide might be handy http://partnerweb.vmware.com/programs/vsan/Virtual%20SAN%20Hardware%20Quick%20Start%20Guide.pdf

Also Duncan Epping has some pretty interesting articles on vSAN, try visiting his blog Yellow Bricks

I wish you good luck!