New ESX Platform - Opinions & Configuration queries.

[spoon_]

Hello.

I have been given the task to write up the draft of new VMware platform. Design document that is.

Basically my employer is seeking to build a platform to host his Information System.

I have to admit straight away - I never done anything like that before and looking for ideas - good or bad, doesn't matter.

What do you guys think document like that should contain?

Hardware to be used, servers:

3x
Dell PE R610, E5530, 2.4GHz Quad CPUs
64GB of RAM
2 x 146GB SAS, RAID1
PERC H700
2 QUAD Intel NICs
All three servers are identical.

Switches, dedicated to the above server:
2 x 24 port Cisco 3750 + 4 SFPs.

Storage:
Existing NetApp SAN array, S500 if I remember correctly, have to confirm this.

Hypervisor:
VMware VSphere Advanced Edition

vCenter server:
As a VM.


Requirements for this platform we have receive from the vendors state that there will be 25-30 VMs running, Windows Server 2008 R2.

Breakdown, still subject to change:

Citrix Servers x 4
Citrix Test Server x 1

DMZ Web Servers x 3

Internal Web Servers x 6

Database Servers x 3

Management Web Server x 1
Management Besiness Server x 1
Management Database Server x 1
Management Test Server x 1

CRM Web/App Servers x 2
CRM Database Server x 1
CRM Test Server x 1

Test Server x 1



What do you guys think?

Does all of this make sense?

Any potential issues with the hardware/software combination and/or lack of horse power?

Any input is gonna be much appreciated!


Thank you,



~Adrian

[spoon_]

I guess we haven't got any project managers lurking around?

[abaxas]

vCenter server:
As a VM.

Danger Danger!

Spend the extra grand on 2x crappy servers to do this.

[spoon_]

Did you have bad experience with vCentre as a VM?

Everyone seems to to be slit between going virtual or physical...

[MaddAussie]

We are running it here as a VM on are small test cluster system and it seems fine

[Moby-Dick]

Danger Danger!

Spend the extra grand on 2x crappy servers to do this.

I dont agree with you I run an Large ESX estate with 1000+VM's and my VC, VC SQL box and VUM box are all VM's.

having just done my VCAP exams , I can confirm its also a VMware recommended practice.

You'll possibly need to supply a bit more information to get more than an "it depends" answer on if you will be able to support that workload on the planned environment. How many Sockets of CPu are you running on each host ?

Are you wanting to run ESXi or Full fat ? you coudl posisbly save some money and go for a USB/SD card install.

You have a fair few DB boxes on there - what disks are they currently working on ? don't skimp on the I/O connectivity.

Server 2k8 R2 is quite RAM hungry - so you'll end up with a minimum of 4GB RAM on each VM , which isn't the end of the world by any stretch - with a minimum of 104GB RAM committed you have sufficient host overhead. how this would leave you with 24GB to allocates ( note I'm assuming you are planning on resources should you loose a Host and still be able to run at full capacity, if you had a host fail while you are patching another one , you'll have some fun )

[spoon_]

You'll possibly need to supply a bit more information to get more than an "it depends" answer on if you will be able to support that workload on the planned environment. How many Sockets of CPu are you running on each host ?

3 hosts, 2 Quads each, 64GB of RAM each.

Are you wanting to run ESXi or Full fat ? you could possibly save some money and go for a USB/SD card install.

Would there be much difference between the two? If so in favour of which flavour?

You have a fair few DB boxes on there - what disks are they currently working on ? don't skimp on the I/O connectivity.

These DB boxes are gonna be brand new installations, the whole information system will be put in place in 1-2 months time. Nothing exists right now.

Server 2k8 R2 is quite RAM hungry - so you'll end up with a minimum of 4GB RAM on each VM , which isn't the end of the world by any stretch - with a minimum of 104GB RAM committed you have sufficient host overhead. how this would leave you with 24GB to allocates ( note I'm assuming you are planning on resources should you loose a Host and still be able to run at full capacity, if you had a host fail while you are patching another one , you'll have some fun )

The exact load figures given by the vendor allow growth so I'm assuming there is some capacity left across 3 hosts, in case 1 goes down 2 hosts will be at full stretch [that's what I think]

Vendors of the application are willing to sacrifice their load balance web servers if need be to decrease potential load on the host for short period of time.

Have you done any design documents before Moby?

[abaxas]

I dont agree with you I run an Large ESX estate with 1000+VM's and my VC, VC SQL box and VUM box are all VM's.

having just done my VCAP exams , I can confirm its also a VMware recommended practice.

You'll possibly need to supply a bit more information to get more than an "it depends" answer on if you will be able to support that workload on the planned environment. How many Sockets of CPu are you running on each host ?

Are you wanting to run ESXi or Full fat ? you coudl posisbly save some money and go for a USB/SD card install.

You have a fair few DB boxes on there - what disks are they currently working on ? don't skimp on the I/O connectivity.

Server 2k8 R2 is quite RAM hungry - so you'll end up with a minimum of 4GB RAM on each VM , which isn't the end of the world by any stretch - with a minimum of 104GB RAM committed you have sufficient host overhead. how this would leave you with 24GB to allocates ( note I'm assuming you are planning on resources should you loose a Host and still be able to run at full capacity, if you had a host fail while you are patching another one , you'll have some fun )

Up to the point of catastrophic failure, especially when you run into datastore problems.

Also terrible for DR as you are controlling yourself!

Just buy the extra machines and be done with it.

[Moby-Dick]

and have a single point of failure when those servers fail ? I'd rather have the benefits of virtualising my own systems thanks

Yeah I've done a couple and have just got to start the one for my VCDX application.

[abaxas]

and have a single point of failure when those servers fail ? I'd rather have the benefits of virtualising my own systems thanks

Yeah I've done a couple and have just got to start the one for my VCDX application.

That is why you have two

[spoon_]

Yeah I've done a couple and have just got to start the one for my VCDX application.

What did you include in your design doc?

Probably what I'm looking here is some sort of table of content kinda thing as I'm not sure what to include and what not - don't want to be too granular and at the same time don't want to miss anything important.

[Moby-Dick]

Have a look at this presentation by a Dutch friend of mine; it should give you some good inspiration ( it did for me )

http://www.gabesvirtualworld.com/pre...are-vsphere-4/

[Moby-Dick]

That is why you have two

and do what with them ? run VCHB ? I'm sure there's a spare 10k in the budget for that.
seriously if I can mitigate the risk of running VC as a VM to a multi-billion pound business, I think I can justify that same level of risk to spoon's boss

[abaxas]

and do what with them ? run VCHB ? I'm sure there's a spare 10k in the budget for that.
seriously if I can mitigate the risk of running VC as a VM to a multi-billion pound business, I think I can justify that same level of risk to spoon's boss

Enjoy your vmotion contention on device failure and inability to do anything on datastore failure.

Wont happen often, but you'll lose your job when it does.

[Moby-Dick]

on which device ? if I have a host failure, Ha will still work. I may loose ability to vmotion for a couple of minutes.

if I loose a datastore , which imply's I've probably lost quite a few VM's, its restore form backup time.
Using Veeam I can have a VM back up (without virtual center) in about 15 minutes. how fast can you rebuild the physical servers ? I'd say given appropriate management sign off no one has to loose their job. In the design given there isn't any DR at all, so a site wide failure is going to leave trousers down all around.

I'd still like to know what availability solution you'd be using.

[spoon_]

How would you guys tackle network configuration given that each server will have two quad NICs?

There are two 24 ports Cisco Catalyst switches dedicated to this task as well.

Obviously iSCSI and VM traffic will be separated and redundant, anything else I need to keep in mind?