Vmware view Datastore access and usage for linked clones

[chriswu11111]

Hi Guys please see attached picture to get a clearer picture of what i'm asking. Again just to state i'm new to vmware view so please bare with me.

I have been provided this solution by a supplier and i just wanted to ask if the data storage can be used by view manager. I will need to support 61 virtual desktops across these 3 servers.

Server 1 = left most server in picture

Server 2 = middle server in picture

Server 3 = right most server in picutre

If you look at the picture server 1 and 2 which is the left most server and the middle server both will have view manager installed on it and hence obviously they can have the linked clones provision the virtual desktops on either servers (I am going to run windows NLB on it to do this).

But the 3rd server (right most server in the picture) doesn't have view manager on it and it is a local storage to the third server. Is it possible to publish desktops on that datastore?

I assume server 1 and 2 are not able to see the datastore on server 3 to be able to provide virtual desktops on it.

Can i assume my assumptions are correct?

If not and this is possible can anyone provide some pointers on how to do this. Please not again for reference server 3 is local storage like server 1 and 2 and is not a san / nas / nfs / iscsi etc.

thanks for your help everyone.    

[gunnarb]

This design is not correct for supporting 61 VMs.  If server 3 goes down the entire View cluster goes down, if 1 or 2 go down you can only support 40 users.  This would not be an acceptable design by anyone's standards. A couple things I would HIGHLY recommend, 4 hosts for starters, provision 20 VMs per host giving you a total of 80VMs, or 60VM with the ability to have a host fail and have minimal impact to your users.  Second, if local storage is a requirement, that's fine... for desktop VMs only.  I would not have any of the server components running on local storage, period.  A seperate cluster could be created that hosts the server VMs.  If a seperate cluster is not possible, then I would spend some money and get the cheapest SAN you can find and have the server VMs sit on the shared storage.  This way if a host goes down HA will kick in and at least the servers will come back so that your View envirnonment doesn't go down like a deck of cards.

To answer your question it doesn't really matter what your VCS can see, it matters what vCenter can see.  So in your world your vCenter will see three hosts, during pool creation you'll select the host and it will create a replica on the same host and from that replica multiple LC VMs will be created.

Gunnar

[chriswu11111]

Thanks Gunnarb!!

Sadly we only have budget for this kind of implementation.

Can you point me on how to set up the 3 hosts to have 20 desktops on each?

I can understand the logic if i use NLB on hosts 1 and 2 and that it can see both local storage for the virtual desktops as linked clones.

How would i get it to see the storage for host 3 to put virtual desktops on there as linked clones as well?

thanks again for your help

[gunnarb]

Sorry man, not to be a jerk but this is such a bad design I'm unwilling to assist as it is designs like this that fail horribly and give the users and management a bad taste for VDI, which negatively affect the entire community.  I know how much of a jerk this makes me sound like and I appolgize.  I have just worked with too many deployments and I know the ones that are bound to fail.  The only help I can send you is to HIGHLY recommend you don't deploy using this design.

Gunnar

[chriswu11111]

Hi Gunnarb,

I understand your thoughts and thank you for your concern. I've raided the system as raid 10 and got enough hard drives for hot spare so we should be good for protection anyway.

In light of this, should i have 4 hosts, and 2 of them being view managers and the other 2 as vcomposer / vcenter redundant systems how does one set up vmware view to be able to deploy to the other 2 local storage which are not sitting on view manager?

[pwynne]

Hi,

I agree with gunarb that this is a flawed design and I would drop the provider of the solution as a starting point. Raid 10 is a good option for your local disk deployment. If you have enough CPU and Memory in your hosts why not drop host 3 and convert it to a NAS node with some 15K SAS disks if you have them. You could run all your VDI''s on local disk and place your management VM's in a resource pool on either ESX Node. You would get your HA at host level without having more than 1 connection server?

Just one option!

Rgds,

Paul.

[chriswu11111]

Thanks for your options guys.

Just for my understanding as im still very new to vmware so please excuse the silly question.

Should host 3 die as per your examples how does that take down the whole view infrasture?

As host 3 is only vcenter and composer I would ahve thought the whole purpose of composer would just be to create linked clones?

So even if host 3 was dead people could still gain access to their desktops hosted on hosts 1 and 2?

The pool we use will be floating pool so will this mean that the users are ok?

The would be hosts 1 and 2 and 3 will be able to host 61 users each should any of them die.

The host spec would be 84gb ram and 1 x 6 core processor.

Thanks guys, Its more for my understanding why this solution doesn't work.

[pwynne]

You lose all your management capability. the pools will stay up but connectivity from the outside will be lost as your security server talks to the connection server which uses your VCenter instance to talk to AD. If you continue with this model you need to keep the VCS and VC server together!

whats the spec of your VDI's?

How much ram is assigned?

How big is the replica disk for each vdi?

How many disks make up your raid10 pool and what spec? these are considerations for your hosts density capabilities.

For example: 8 x 15K SAS disks configured as Raid 10 will allow for 800 IOPS taking penalties into account. the raid controller cache should give a little bit more performance before you see any kid of latency.

I have placed 80 linked clones on a Del R710 with 96GB ram,2 x Xeon 5670 CPUs all on local disk and found it works perfectly so you could take Host1 and put the management VM's (VCS,VC,DC etc...) on it with maybe 10 or 20 clones, then on the second host place the remaining LC's using the third host as your shared storage. raiding on the third host will give you redundancy at a disk level so you just need to ensure the host is running on a clean supply to avoid electrical damage..

[chriswu11111]

Our hosts are:

HP ProLiant DL380 G7 Intel Xeon X5650 CPU, 84 Gb RAM, 11x 300Gb
HDD 15k, Redundant Power , 4Gb SD, RAID upgrade to 1Gb BBWC
4x 1Gb NIC. SAS Expander, iLO Adv

So RAID 10: 10 x 300GB drives above.

The VDI: windows xp, 2gb ram, 15gb drives.

Non persistant disks, floating pool, 3 hosts to support 61 users. so 20 users per host

We won't be providing features like local mode, or remote access from home so the security server won't be needed. In addition a juniper vpn appliance box now supports vmware view in its hardware so during the test phase we didn't need a security server the vpn appliance is able to talk straight to the view manager box without needing a security server.

In addition when you say management i assume you mean management of the ESXi server's and not management of the view structure as view manager should be still capable of running and managing user sessions and pools that have already been made even when composer and vcenter is down. I understand that composer can't be used to make new pools and desktops but i will only be using one floating pool anyways.

For your solution of having 2 hosts and 1 host as a san or nas function what will be the redundancy in this since if this goes down that is also one single point of failure if all the view manager, composer, vcenter etc are all put on that.

thanks again!

[pwynne]

Hi,

Thats correct that you would have a single point of failure for the server that would act as a NAS however right now if server 3 goes down you are crippled.

My solution can offer you VM level resiliance through ESX HA and disk level protection through a sizeable raid set on your server 3 acting as a shared storage resource, you would not have protection at a hardware level and that is something you dont have right now really. You can go a step further and add a VM as a file server and replicate all of your key data to the vm homed on host 1 or 2.

Really this would be a way of improving the design you have. I really think its worth purchasing a storage array if you can get budget. Dell Equalogic 4x or MD3200i would be good solid storage at a good price.

FYI: I like what you are doing with VDI on local disk its emerging as a solid cost effective delivery platform now for VDi in many markets.

Hope this was a help

Paul.