cluster networks microsoft slides ppt, Assignments of Computer science

Download cluster networks microsoft slides ppt and more Assignments Computer science in PDF only on Docsity!

Click to edit Master subtitle style Microsoft Virtual Academy

03 | Cluster Networking

Elden Christensen | Principal Program Manager Lead

| Microsoft

Symon Perriman | Vice President | 5nine Software

• (^) Cluster Network Infrastructure fundamentals
• (^) Cluster Network Design Planning
• (^) Cluster Network Configuration Options Module Overview

Health Monitoring of Nodes in the Cluster

• (^) Failover Clustering conducts health monitoring between nodes to detect when servers are no longer available
• (^) When servers are unresponsive clustering takes recovery action
• (^) Unicast in nature and uses a Request- Reply type process for reliability and security - (^) Not just a basic ping You there? Yes You there? Yes

Network Health Detection

• (^) Clustering does full mesh monitoring between all network interfaces between all nodes
• (^) Health monitoring:
  • (^) Nodes exchange heartbeats every 1 second

(configurable)

• (^) Nodes are considered down if they do not

respond to 5 heartbeats (configurable)

• (^) Nodes are removed from cluster membership if they exceed thresholds
• (^) Communication over port 3343

Viewing NetFT Virtual Adapter

• (^) NetFT is a virtual network adapter
  • (^) Visible in Device Manager and with IPConfig /all
• (^) Completely self configuring
  • (^) Media Access Control (MAC) address is self-generated based on a hash of MAC address of the first enumerated (by NDIS) physical NIC in the cluster node
  • (^) NetFT self-configures an APIPA (Automatic Private Internet Protocol Addressing) address
  • (^) No manual user configuration required

NetFT Architecture

• (^) NDIS 6.2 miniport virtual adapter
• (^) Supports Receive Side Scaling (RSS)
• (^) Network fault tolerance for TCP and UDP across routed network connections - Each link independently monitored
• (^) Supports IPv4 and IPv
• (^) Built in route failure detection
• (^) IP over UDP/IP Tunneling
  • Virtual adapter tunnels over physical adapters
• (^) Clustering leverages port 3343
  • (^) NetFT uses UDP 3343
  • (^) ClusSvc uses TCP 3343

NDIS

IP

TCP UDP UDP

NetFT NIC1 NIC

ClusSvc

Cluster Network Discovery

• (^) Cluster uses exactly one IP per Subnet per NIC
  • (^) Cluster ignores other IPs from the same subnet configured on the NIC
  • (^) Cluster ignores other NICs & associated IPs from the same subnet
• (^) Each NIC per Node will be part of exactly one Cluster Network - (^) Cluster will use prefix matching to determine the set of Cluster Networks - (^) Cluster has built-in resilience to use IPv4 or IPv6 per NIC (prefix must match)
• (^) Cluster will use an IP from different Subnet from another NIC
  • (^) Cluster will ignore other IPs configured on that NIC for subnets already discovered

Cluster Network Discovery: Example Same Subnet NIC 2 Same Subnet NIC 2

Cluster Network 1 Cluster Network 1 NIC 2 Ignored By Cluster NIC 2 Ignored By Cluster Same Subnet NIC 1 Same Subnet NIC 1

Cluster Communication

• (^) Clusters exchange three types of communication: Network Health Monitoring
• (^) Heartbeats are sent to monitor health status of network interfaces
• (^) Are sent over all cluster enabled networks Intra-cluster Communication
• (^) Database updates and state synchronization that are sent between the nodes in the cluster
• (^) Example: When creating a new resource the cluster database must be updated on all nodes
• (^) Are over a single interface CSV I/O Redirection
• (^) Metadata updates to files
• (^) All I/O in failure scenarios
• (^) Over same network as intra-cluster communication
• (^) Over a single interface
• (^) Can leverage SMB multi- channel to stream over multiple interfaces

DEMO

Microsoft Virtual Academy

Manage Cluster Networking

Network Bandwidth Planning

• (^) Lightweight (only 134 bytes)
• (^) Sensitive to latency
  • (^) If cluster heartbeats become blocked by a saturated NIC, this could cause nodes to be removed from cluster membership
• (^) Bandwidth not important, but quality of service is Heartbeats
• (^) Lightweight
  • (^) Traffic varies by workload, in general infrequent on running stable File / Hyper-V clusters. Heavier on SQL / Exchange clusters
• (^) Clustering is a distributed synchronous system, latency will slow down cluster state changes (such as failover)
• (^) Bandwidth not important, but quality of service is Intra-Cluster Communication
• (^) Metadata updates
  • (^) Lightweight and Infrequent
  • (^) Latency will slow down I/O performance
    • (^) Yes, network performance will impact storage I/O performance!
  • (^) Quality of service most important
• (^) Failure scenarios / asymmetric storage configurations
  • (^) All I/O is forwarded via SMB over the network
  • (^) Network bandwidth is most important CSV I/O Redirection Key Take-away: Primary design consideration for cluster communication is ensuring quality of service

Traditional Network Configuration Guidance

What we have recommended for the last decade…

• (^) At least 2 independent networks
  • (^) Public / Client Network
    • (^) IPv4 - Static or DHCP assigned address » APIPA (aka. autonet) 169.254.x.x addresses not supported
    • (^) IPv6 Stateless address autoconfiguration (SLAAC) » (^) Note: DHCPv6 not supported for cluster IP Address resources
    • (^) Default gateway (routable)
  • (^) Intra-Node Communication
    • (^) IPv6 (preferred) or IPv » (^) IPv6 Linklocal (fe80) works great
    • (^) No default gateway (non-routable)
    • (^) Separate physical network

This still applies, but we also support converged networking

Are Separate Networks Really Needed? Required?

• (^) No – It is not required to have 2 separate networks
  • (^) Clustering does support a converged networking model
  • (^) Validate will generate a Warning to alert you of a potential single point of failure
    • (^) Validate is not NIC Teaming aware Recommended?
• (^) Yes – It is recommended to have redundant network communication between nodes
• (^) Sort of… let’s talk about what really matters and converged networking (next slide)

Resiliency

• (^) In a highly available system you want to avoid any single points of failure
• (^) Many ways to accomplish network redundancy - (^) Multiple independent

networks

Quality of Service

• (^) Cluster heartbeats are lightweight, but sensitive to latency - (^) If cluster heartbeats can’t get

through… this can be falsely

interpreted that nodes are

down

• (^) Many ways to accomplish network quality of service - (^) Multiple network cards Converged Network Considerations