5. Implement High Availability Flashcards
Hyper-v Replica
Allows you to create highly available VMs locally or on a remote site through asynchronous or synchronous replication of storage.
Hyper-v replica works off of snapshots so after the initial replication the only thing that needs to be sent over the network is the checkpoint.
kerberos
replica traffic will not be encrypted and the servers must be joined to the same or trusted domains.
requires constrained delegations to be established in active directory.
do not need to be signed into the server you are migrating from.
certification-based authentication
uses certificates for authentication for encrypted replica over the network. you can either self-sign certificates or have them signed externally.
Live migration
A feature that makes it possible to move a virtual machine from one hyper-v host to another while it is running with almost no interruption of service.
this is not a replacement for hyper-v replica because it does not move the virtual machines data files.
Simultaneous live migrations
allows you to specify how many live migrations the server can perform at the same time based on the bandwidth and traffic levels on the server. the default is 2 live migrations at a time.
credssp
an authentication protocol that does not require constrained delegations to be established.
requires that you be signed into the server you are migrating from.
shared nothing live migration
allows you to move vms that do not have shared storage by moving the storage alongside the vm.
basically a storage migration then a live migration.
must be members of the same or trusted domain.
the source and domain must be using the same processor family amd vs intel
minimum 1gb network
virtual switches should be configured identically between servers.
Storage migration
the process of moving vm storage to another vm.
Failover clustering
a group of two or more computers, physical or virtual, and runnin the same application that functions on a single entity to provide high availability, scalability, and fault tolerances.
64 computers
8000 virtual machines
1024 vms per node.
Single domain cluster
all of the servers in the failover cluster are joined to the same domain.
Multi-domain cluster
a failover cluster with servers running in multiple different domains.
workgroup cluster
a failover cluster of computers not joined to a domain at all.
SQL would work well in a domainless cluster.
file servers would not work well in a domainless cluster.
Quorum
a voting system in a cluster to prevent a split-brain scenario. The quorum keeps one side of the cluster up because it has a majority vote.
Disk witness
a dedicated disk in the cluster’s shared storage that contains a copy of the cluster database.
file share witness
an SMB file share on a windows server with a witness.log file containing information about the cluster.
cloud witness
a blob stored in azure that is designed for stretch clusters split among multiple data centers and remote sites. the purpose of the cloud witness is to remain independent of all of the data centers.
Dynamic Quorum Management
designed to function so that when a node leaves a cluster its vote leaves with it so that the cluster cna stay up without taking itself offline. allows a cluster to continue functioning until all of the nodes have failed.
Client Communication
Clients access to the application running on the cluster are the highest priority and this is usually the default shared network.
Cluster communication
cluster communication is the next highest networking priority. if the cluster cannot communicate with itself and register heartbeats the cluster will go down.
storage traffic
iSCSI and other types of storage traffic should be separated from all other types of traffic on the cluster network.
cluster storage types
Fibre Channel
SAS
iSCSI
cluster aware updating
allows you to update clusters with minimum downtime.
remove all active roles from a node then update it. bring the node online and do the same to the next one until you reach the final node.
self updating mode
a node in the cluster has the CAU role and starts to update all of the nodes in the cluster. when the CAU reaches itself it fails over and a new node assumes the role and updates the old CAU.
remote updating mode
a computer outside the cluster is configured to function as the update coordinator from this computer an admin can manually start the update process and walk through each of the steps.
cluster OS rolling upgrade
makes it possible to upgrade OS versions in a cluster. this works like cluster-aware upgrading we are going to remove a node and perform a clean install on the next os version and migrate the roles back to the server and rejoin it to the cluster. we are going to do this with each node in the cluster then we are going to update the cluster functional level. you cannot go back to a previous version of windows server once you have upgraded you would need to build a new cluster.
CSVFS
Cluster shared volume file system
regular ntfs file systems were not made for clustered systems. only one node can access the metadata or there would be corruption and your data would be ruined. this is the purpose of a CSVFS where one node modifies the metadata and has complete control over it and other nodes can edit the actual files.
administrative access point
is the cluster object name that is created in active directory when the cluster is created.
Active Directory Detached Cluster
a cluster without an active directory object even though the cluster nodes are joined to a domain. instead the detached cluster will use the dns as the administrative access point
new-cluster cluster1 -node node1,node2 -staticaddress 10..0.0.1 -nostorage -administrativeaccesspoint dns
Scale-Out File Server
a clustered role that is designed to provide highly available storage for applications such as hyper-v and SQL. this means that storage is available to all cluster nodes at the same time. active active or dual active
Guest Cluster
is a cluster that consists entirely of virtual machines running on a single hyper-v host.
Node Monitoring
Clusters can monitor resources like storage subsystems, network connectivity, and the clustered application itself.
application migration
allows an application to failover to other vms
Host Availability
you can create a guest cluster from virtual machines located on different hosts allowing a host to fail and still preserve your workloads.
nested clustering
a guest cluster within a cluster. this is good for creating a cluster made up of guest clusters.
storage spaces
allows you to add data storage provided by multiple physical disks to a storage pool. you can then combine this storage space with failover clustering to create a solution that is highly available to both hard drive and server failures.
supports parity and mirroring
unmonitored
indicates that a vm owning a role is not being monitored by the cluster service.
isolated
indicates that the node is not currently an active member of the cluster but is still in possession of a role.
quarantine
indicates a node that has been drained of its roles and removed from the cluster for a specified amount of time after having left and reentered the cluster too many times in the previous hour.
resiliency level
1 - enables the use of the isolated state only if th node supplies a known reason for disconnecting.
2 - enables the free use of the isolated stat and allows the node time to recover. (default)
VHDX
allows files created on a shared storage infrastructure to be shared between virtual machines in a gues cluster. for backwards compatability at this point. 2016 uses a vhd set.
VHD Set
contains a vhds file for the metadata and an avhdx that contains the actual data. vhd sets provide online resizing capabilities and supoort for host based backups that are not supported by vhdx files.
Storage Spaces Direct
allows you to create storage spaces from the internal storage of servers without the need for external storage solutions.
data center only feature
minimum of 6 drives with 4 being hard disks and 2 being solid state
drives cannot be partitioned before setup of s2d
10gb connections between the nodes or servers
uses smb
software storage bus
a logical network that connects the local data drives inside them.
SMB Transparent Failover
Enables a client session to be transferred from one cluster node to another without interruption.
SMB Scale-out
enables share to be accessible by clients from all nodes in the cluster simultaneously.
SMB Multichannel
enables file servers to combine the bandwidth from multiple network adapters to increase bandwidth and fault tolerance.
SMB Direct
uses RDMA to perform direct memory to memory data transfers between the remote systems this decreases processor load for the host. requires smb3.0 for both the client and the server.
SMB encryption
provides end to end aes encryption between servers and clients using smb 3.0
Vm monitoring
a cluster is capable of monitoring specific services on the virtual machines and reporting when a problem occurs and taking action that you can configure.
the vm must be in the same domain
the firewall must be set up to receive inbound rules
failover
when a role running on a cluster node can no longer continue to run the cluster moves it to another node.
failback
when a cluster moves the role back to its original node after the problem that caused the node to failover is resolved.
stretch cluster
a cluster with its nodes deivied among different sites often in different cities.
Site aware failover cluster
a cluster that contains fault domains the cluster uses the fault domains to determine its behavior during failovers and role transfers.
allows you to set primary and secondary sites
fault domain
a set of hardware components that share a single point of failure.
failover affinity
nodes have a preference for their site they will not failover to another site until all nodes at their site have failed. they have affinity for their site./
crosssitedelay
specifies the amount of time between heartbeats sent to nodes of different sites.
crosssitethreshold
specifies the number of missed heartbeats that must occur before a node at a different site is considered to have failed.
node fairness
attempts to balance the distribution among the nodes. works by evaluating the memory and cpu loads on each node over time attempting to identify those that are overcommitted.
auto balancer mode
0 - node fairness diabled
1 - laod balancing occurs when a node joins a cluster
2 - load balancing occurs when a node joins a cluster and every 30 minutes
auto balancer level
1 - low. migrates vms when the host is more than 80% loaded
2 - medium. migrates the server when it is more than 70% loaded
- high. migrates the server when it is more than 60% loaded
Export/import vm
effectively allows you to clone a vm.
generate new id - generates a new id for the vm allowing you to run it and the original
Network load balancing
distributes traffic load among many computers enabling the cluster to service more clients at once.
2 to 32 hosts
best suited for stateless applications
must be on the same subnet
must have static ip addresses
all hosts should have an identical user account
Stateless application
an application that does not store client data for use in one session with another session
a great example of this is html
Failover clustering vs load balancing
load balancing is for hosts failover clustering is for nodes.
Convergence
when adding or removing hosts from a cluster the NLB will evaluate the current cluster membership and determine how client requests should be distributed among the hosts.
DNS round robbin
allows dns servers resolving the cluster name to supply different ip addresses to successive requests.
Unicast
replaces the hardware mac address with the cluster’s virtual mac address causing traffic to go to every host in the cluster and every port. with unicast, your hosts cannot communicate with one another.
Multicast
adds a second mac address to the network interface of each of your hosts this allows your hosts to continue communicating with one another but can still cause port flooding, “but there are fixes for it”
IGMP (internet group management protocol) multicast
uses igmp to program the switches so that traffic destined for the cluster’s mac address is only forwarded through the switch ports connected to nlb hosts.
instead of doing this you can also create a vlan in the switch
no affinity
your clients do not have an affinity for a particular host meaning their traffic is sent to any host in the nlb cluster.
single affinity
your clients have affinity for a particular host.
network affinity
your cluster recognizes the class c section of an ip address and assigns affinity to the client for a particular server that would best suit them.
Load weight
specifies how much traffic conforming to a rule should be handled by a particular host. multiple host mode
0-100
handling priority
specifies the priority for the hosts handling of traffic for a particular rule in single-host mode.
simultaneous upgrade
bring the entire nlb cluster down and upgrade all the hosts then bring the cluster up again.
rolling upgrade
remove the hosts from the nlb cluster one at a time to upgrade each one then add it back to the cluster.
