VMware VIEW COMPOSER 2.5 - ARCHITECTURE Manual do Utilizador descarregar pdf (Página 22)

EMC Infrastructure for Virtual Desktops Enabled by EMC VNX, VMware vSphere 4.1,

VMware View 4.5, and VMware View Composer 2.5

Processor

The server used in this solution has two quad-core Intel Xeon 5500 processors. The average

CPU load during the test is 9 percent. Therefore, we can run approximately 10 virtual

machines per core. One host can run 2 × 4 × 10=80 virtual machines. The Intel Nehalem

architecture is very efficient with hyper-threading and allows 50 to 80 percent more clients.

This means it can run 1.5 × 80=120 to 1.8 × 80=144 virtual machines per host.

While using linked clones, up to eight hosts are allowed in a cluster. Leaving one node as

failover capacity, with seven hosts, we can run 144 × 7=1008 virtual machines. One cluster

can host 1,000 virtual desktops. Without considering the Intel Nehalem features, the cluster

can host 80 × 7=560 virtual desktops. To host 2,000 virtual desktops, we need two to four

clusters, which are about 128 to 256 processors in total. In a non-VDI environment, deploying

2,000 desktops would require 2,000 processors.

With hyperthreading, we are able to host 1,000 VMs per cluster and without hyper threading,

we are able to host only 500 VMs per cluster. Thus, we are able to double the number of

hosts per cluster when using hyper threading. In our solution, we use hyper threading with

three clusters; one with 1,000 users and other two with 500 users each. The 500-user cluster

has extra room for processor intensive workloads.

Table 6 provides a summary of virtual machines per core.

Table 6. Virtual machine per core

Case

Complete Cluster

Cluster with one node down

1000-user cluster

16 VMs per core

18 VMs per core

500-user cluster

8 VMs per core

9 VMs per core

Memory

One Windows 7 virtual machine is assigned 1.5 GB of memory. Without using VMware

vSphere 4.1 features, it would require at least 9×8×1.5=108 GB to 18×8×1.5= 216 GB per

host. VMware vSphere 4.1 provides features such as Transparent Page Sharing, ballooning,

compression, recognition of zeroed pages, and memory compression that enable us to allow

over-committing the memory to obtain a better consolidation ratio.

During the baseline workload, we observed about 540 MB used in active memory. The

memory overhead was 179 MB, the hypervisor used 578 MB for the 48 GB host and 990 MB

for the 96 GB host, and the service console memory was 561 MB. Based on this workload, we

require:

((9×8×(540+179) +578+561)/1024) = 52 GB to ((18×8×(540+179) +990+561)/1024) = 103

VMware vSphere uses the above-mentioned features before it uses the swap memory. The

FAST Cache on EMC’s VNX series storage platform does provide better response time

compared to swapping the memory to SAS disks. Another option is to consider having SSD on

each host to host the vSwap. This may impact vMotion and also adds complexity to the

environment. It is, therefore, advantageous to have the swap served by the FAST Cache on the

EMC array.

1 2 ... 17 18 19 20 21 22 23 24 25 26 27 ... 62 63

Comentários a estes Manuais

Sem comentários

VMware VIEW COMPOSER 2.5 - ARCHITECTURE Manual do Utilizador Página 22

Comentários a estes Manuais

Manuais e produtos relacionados com Software de gestão do sistema VMware VIEW COMPOSER 2.5 - ARCHITECTURE