Designing a solution to have 100 VM on 4 hosts

[Renmo]

Hello Community,

since 60 servers will be migrated from p2v and another 50 VM will be installed in the near future i was wondering which host specs would you go for ?

memory varies between hosts as well as CPUs

i'm planning  to suggest 4 servers, each with 128 GB Memory and two Xeon® X5675 (3.06GHz/6-core/12MB/95W), what woudl you think about it ?

Network wise i will have FLEX10 and 10 g NIC, and since it's a blade solution they will be connected to SAN

i know the info i have given are brief, please feel free to ask me for any clarification

Thanks in advance

[AndreTheGiant]

With 100 VM you need to put some attention on the vRAM (also if you plan to go on vSphere 5 ) and ALSO on storage side.

Have you estimated how much RAM are you actually used? and which kind of vRAM you need on the new VMs?

Andre

[Renmo]

Hello Andre,

I had a feeling that you would be the first one to answer ! actually this is the scenario, the CP was done for 120 machines and the memory requirement for them varies, and only 50 from them will be migrated to V and another new 50 servers will be created.

so i have made an assumption like the below:

-      Around 460 GB memory is distributed between 120 servers and

460/120= 3.8 GB memory if it was distributed equally but since 50 servers will be migrated only we will assume that important ones will be virtualized will do the following assumption

8 servers with 8 GB         8*7=56GB

6 servers with 6 GB         6*8=48GB

35 servers with 4 GB       4*35=140GB

Total = 244 GB

And if the same calculation is done for the new virtual machines that will be implemented

8 servers with 8 GB         8*7=56GB

6 servers with 6 GB         6*8=48GB

35 servers with 4 GB       4*35=140GB

Total = 244 GB

Total for migrated and new servers is 488 GB minus 50 GB Swap

488/4= 122 GB per Host (for the memory configurations)

And since it’s recommended to have n+1 for HA and DRS will add one extra server

as for the CPUs and Vcpue considerations:

each Vcpu can have around 3 VM so with 2 socket CPU 6 cores it will be like this ( 2 cpus * 6 Cores = 12cores)

and 12 cores * 3 Virtual machines = 36 VM and if i assign two Vcpu i will have around 20 VM

so based on that i was planning to suggest 5 hosts, each 128GB plus one host for the HA and DRS

what do you think about it ?

and no we are going to use 4.1

[Renmo]

And also, forgot to ask you about the storage, i need to decide how much space is going to be needed, and was thinking of the following

70GB for C: for each machine, and 30 for 😧 for windows machine

for linux i could allocate 100GB for each   so

100GB*100 VM = around 10 TB ?

i have read about calculating the I/O but not really sure how that's done or how it's calculated and what needs to be taken under considerations

[weinstein5]

Your calculations look good - they should work but I am always concerned about memory so I would suggest 192 GB per server instead of 128 GB - but with your calcualtions and ESXi ability to do Transparent Page Sharing and adding an extra machine it should be fine.

In terms of storage it looks good but also take into consideration the space for the per VM VMkernel swapfile which is created when the VM is powered on - the size of the file is equal to the difference of the memory assigned to the VM and its reservation - so in a worst case scenario you will need disk space equal to the amount of the memory assigne to the VMs or 488 GB - also do not forget space for templates and growth - 

[Renmo]

Those were great points to put under consideration, how much do u suggest to assign for the templates and why ?

and in regard to SRM ... i might have suggest a solution for it but i'm not very sure of the DR site requirments details, but any points that could pop up in your mind in regard to designing the SRM would be appreciated , i'll try to give mroe details about the requirments soon

thanks a lot

[AndreTheGiant]

Templates could be really few if you can standardize your OS.

About SRM, this could really depend on your storage type (actually SRM can ONLY work with storage replication).

So, if you plan to implement it, choose the storage also with this aspects.

Then most of SRM design is based on how do you create the LUNs (and of course the datastores)... but the hosts ans storage sizing (on the production) is not affected.

Andre

[weinstein5]

As Andre indiciated SRM requires SAN to SAN replication - if you can wait until the release of vsphere 5 and SRM 5 - SRM will no longer require SAN to SAN replication - they should be out by the end of next month

[FredPeterson]

Don't forget that many physical machines have more memory then they ever use or will ever use.  So its likely you have many 4GB machines that only really need 2GB and possibly some Windows 2000 machines with 2 that only need 1 etc.