Hardware Management - Deploying and Managing a Cloud Infrastructure - page 241

Distributed journaling enables recovering of virtual machines in the event of
server failure.

■

Uses large block sizes for virtual disk I/O and sub-block allocator for files and
directories.

■

VMFS3 enables connecting up to 32 hosts.

■

Single volume supports up to 64 TB.

■

Enhanced VMFS performance with volume, device, object, and buffer caching.

■

Summary

The chapter is called “Hardware Management,” and we went into detail about how BIOS
and firmware configuration can be used to sort out incompatibilities and even increase per-
formance through various tweaks. For now, most service providers do not explicitly offer
end-users direct BIOS configuration capabilities, but users can make special requests to the
service provider or make changes through special configurations.
Memory is an important resource because it can allow for more workload to be handled
by the virtual machine. But control in allocating memory should be exercised because excess
allocation can add up, especially if there are multiple instances. Imagine allocating an extra
gigabyte of memory to each of 100 virtual machines, which would be a crazy amount. In
this situation, lean really is mean.
In addition to memory, virtual machines need compute resource. The vCPU resource is
the heart of the VM, but it is the only resource that does not explicitly belong to the VM
because it is only a representation of time slices that are allocated to each instance. The allo-
cation of CPU time is proportional to the number of vCPUs and cores that are assigned to a
virtual machine. This has to be scaled with the type of workload that a VM is subjected to.
If it is a mere database server, it might not need much computing power, but if it is a server
used for analytics, it might benefit from more vCPUs with multiple cores. Hyperthreading
doubles the number of cores that the host has.
If the workload is more network hungry than storage or compute hungry, you need a
good interface with the network, and a single NIC will usually not be enough. The host can
use a single NIC for multiple virtual machines by attaching that NIC to a virtual switch
component, but the network will be shared among those VMs and is really just useful for
workloads that do not demand much network traffic. If the workload is network intensive,
the ideal setup would be a single physical NIC per virtual machine, which would ensure
that the VM has the whole pipe to work with.
When you're selecting host hardware, compatibility has to be taken into consideration.
Although most hardware modules are made to be swappable and interchangeable, incompat-
ibilities will still occur because of the different manufacturing processes and configurations
of the modules. To avoid downtime and performance degradation, it is important that no
inconsistencies and incompatibilities occur between host hardware resources.