Exam Night Mode

The night before the exam

No fluff — just the definitions, comparison tables, and must-memorize points you need to review fast.

Week 1

Cloud Basics

Full lesson

📖 Must-Know Definitions

Cloud computing = on-demand network access to a shared pool of configurable resources.
CapEx: upfront capital spending on assets you own.
OpEx: ongoing operational spending for services you consume.
Rapid Elasticity: ability to scale resources up/down quickly to match demand.
Cloud provides business agility → quick time to market and reduced infrastructure investment.
IaaS delivers infrastructure; PaaS delivers a development platform; SaaS delivers finished software.
Private cloud is dedicated to one organization; community is shared by a specific group.
Digital transformation: using technology to accelerate business and improve customer experience.
Orchestration: automated provisioning and coordination of cloud resources.
Reference architecture: layered model of service, control, virtual, and physical resources.

🎯 Key MCQ Points

The NIST definition keywords: on-demand, shared pool, rapidly provisioned, minimal management.
Cloud is a MODEL, not a single technology.
CapEx = traditional, OpEx = cloud.
Elasticity and speed of deployment are the key cloud advantages.
There are exactly FIVE essential characteristics.
'Measured Service' = metering/billing transparency.
'Resource Pooling' implies multi-tenancy.
Slides use 'Broadband network access' — same as Broad Network Access.
Typical uses from slides: backup, software testing, SaaS, seasonal peaks.
Key benefits: business agility, reduce IT costs, high availability, flexible scaling.
IaaS = most control, SaaS = least control for the user.
PaaS targets developers; SaaS targets end users.
Hybrid = public + private working together.
Community = shared concerns among multiple organizations.
IT transformation drives cloud adoption.
Business challenges: shrinking markets, rising competition, time to market, IoT implications.
IT challenges: data growth, aging tech, poor scalability, shadow IT, financial pressure.
Cloud benefits from slides: agility, reduce investment, improve utilization, reduce management.
Self-service and automation are process-level IT transformation goals.
Control layer sits between service layer and virtual/physical layers.
Private cloud adds pooling, automation, and self-service on owned infrastructure.

⚠️ Common Mistakes to Avoid

Thinking cloud only means 'storing files online' — it includes compute, network, and platforms too.
Confusing CapEx with OpEx in cost-model questions.
Mixing up Rapid Elasticity (scaling) with Resource Pooling (multi-tenant sharing).
Forgetting seasonal peaks as a classic cloud use case.
Saying the user manages the OS in PaaS — the provider does.
Confusing community cloud with public cloud — community is restricted to a group with shared goals.
Treating cloud as only storage — it's the enabler of digital transformation.
Forgetting 'People' as a pillar of IT transformation — it's not just technology.
Equating any on-premise data center with a private cloud — automation and pooling define private cloud.

Traditional IT vs Cloud

Aspect	Traditional IT	Cloud
Cost model	CapEx (buy upfront)	OpEx (pay-as-you-go)
Scaling	Slow, buy new hardware	Elastic, on-demand
Management	You manage everything	Provider manages infrastructure
Time to deploy	Weeks / months	Minutes
Capacity	Fixed, over-provisioned	Flexible, right-sized

Who Manages What

Layer	IaaS	PaaS	SaaS
Applications	You	You	Provider
Data	You	You	Provider
Runtime / OS	You	Provider	Provider
Virtualization	Provider	Provider	Provider
Hardware / Network	Provider	Provider	Provider

Traditional Data Center vs On-Premise Private Cloud

Aspect	Traditional DC	Private Cloud
Provisioning	Manual, slow	Automated, self-service
Resource pooling	Siloed per app	Pooled and shared
Management	Element-by-element	Unified control layer
Elasticity	Fixed capacity	Dynamic allocation

Week 2

Physical Layer — Compute

Full lesson

📖 Must-Know Definitions

Blade chassis provides shared power, cooling, and networking to blades.
HCI converges all infrastructure components into one managed platform.
Virtualization = abstracting physical resources into virtual ones.
Hypervisor/VMM: software layer that manages VMs and allocates hardware resources.
Encapsulation: representing an entire VM as files.
Paravirtualization: guest OS is modified to make hypercalls to the hypervisor.
Time slicing divides CPU time among multiple VMs.
Snapshot: a saved point-in-time state of a VM for rollback.
I/O virtualization: virtualizing storage and network devices for VMs.
Live migration: moving a running VM between physical hosts.

🎯 Key MCQ Points

Blades share power/cooling via the chassis → higher density, less cabling.
Rack servers are independent units.
HCI = compute + storage + network + hypervisor, software-defined, single managed system.
Infrastructure virtualization → hardware; process virtualization → a single app environment.
Type 1 = bare-metal = better performance.
VMM is another name for the hypervisor.
Encapsulation = VM as files → enables easy backup and portability.
Isolation improves security and stability.
Paravirtualization REQUIRES a modified guest OS (hypercalls).
Hardware-assisted uses CPU extensions (Intel VT-x, AMD-V).
Full virtualization uses binary translation, no OS modification.
Time slicing = sharing CPU time across VMs in turns.
Snapshot = point-in-time state, NOT a full backup.
Virtual disk holds OS + data; config file holds hardware settings.
Encapsulation enables snapshots and clones.
Hypervisor abstracts processor, memory, network, AND storage of the compute system.
Portability enables migration of live, running VMs.
Snapshot captures state; clone creates a new independent VM.

⚠️ Common Mistakes to Avoid

Saying blade servers are fully independent — they depend on the chassis.
Forgetting that the hypervisor is a core part of HCI.
Confusing process virtualization (e.g. JVM) with full machine virtualization.
Mixing up Type 1 (on hardware) and Type 2 (on host OS).
Confusing isolation with encapsulation.
Saying full virtualization modifies the guest OS — it does not.
Thinking each VM gets a dedicated physical core by default.
Treating a snapshot as a permanent backup.
Thinking only CPU and memory are virtualized — I/O and storage are too.
Using snapshot as a substitute for backup or clone.

Rack Server vs Blade Server

Aspect	Rack Server	Blade Server
Power & cooling	Self-contained	Shared via chassis
Density	Lower	Higher
Cabling	More cables	Reduced (shared backplane)
Cost (small scale)	Lower upfront	Higher (needs chassis)
Best for	Small / mixed loads	Dense, large deployments

Type 1 vs Type 2 Hypervisor

Aspect	Type 1 (Bare-metal)	Type 2 (Hosted)
Runs on	Hardware directly	A host OS
Performance	Higher	Lower (extra OS layer)
Use case	Data centers / production	Desktops / testing
Example	ESXi, Hyper-V, KVM	VirtualBox, VMware Workstation

Full vs Paravirtualization vs Hardware-Assisted

Aspect	Full	Paravirtualization	Hardware-Assisted
Guest OS modified?	No	Yes	No
Technique	Binary translation	Hypercalls	CPU extensions (VT-x/AMD-V)
Performance	Moderate (overhead)	High	High
Compatibility	High (any OS)	Lower (needs modified OS)	High

Snapshot vs Clone vs Backup

Type	Purpose	Independent?	Note
Snapshot	Point-in-time rollback	No (depends on parent)	Not a full backup
Clone	Duplicate VM	Yes	Full independent copy
Backup	Disaster recovery	Yes	Stored off-system

Week 3

Physical Layer — Storage

Full lesson

📖 Must-Know Definitions

Sector: smallest addressable storage unit on a disk.
SSD: solid-state drive using flash memory, no mechanical parts.
Parity: redundant data used to reconstruct a failed disk.
Write Back: data written to cache first and to disk later.
Object storage: data stored as objects with metadata, accessed via API.
Scale-out NAS: cluster of nodes providing unified NAS capacity and performance.
Storage virtualization: abstracting physical storage into logical pools.
LUN: a logical disk unit presented to a host.
MetaLUN: method to expand LUNs requiring additional capacity or performance.
Thin provisioning: storage allocated as needed, not upfront.
Front end: the interface through which hosts access storage.
Write penalty: extra I/O operations required due to redundancy mechanism.

🎯 Key MCQ Points

Order from large to small: Platter → Track → Sector.
Sector is the smallest addressable unit.
SSD = no moving parts, lower latency.
HDD = lower cost per GB.
RAID 0 = no fault tolerance (striping only).
RAID 3 = single dedicated parity disk; all disks involved in parallel I/O.
RAID 5 tolerates 1 disk failure; RAID 6 tolerates 2.
RAID 1+0 (RAID 10) = mirroring + striping combined.
RAID 6 write penalty: 6 I/O ops (3 reads + 3 writes) vs RAID 5's 4.
Write Through = safer (writes to disk immediately); Write Back = faster (risk on power failure).
LRU evicts least recently used; MRU evicts most recently used.
Cache Vaulting protects write-back cache during power loss.
Watermarking: Idle, High watermark (HWM), and Forced flushing manage cache utilization.
SAN = block-level; NAS = file-level; Object = API/metadata-based.
Databases prefer block (SAN).
Traditional NAS = scale-up; Scale-out NAS = cluster of nodes pooled.
Scale-out NAS = multiple nodes pooled as one NAS device.
Three locations: host, array, network.
LUN masking = access control at the storage array (which host sees which LUN).
Concatenated = capacity only; Striped = capacity + performance.
Thin = allocate on demand; Thick = allocate all upfront.
Front end = host connection; back end = physical disks.
RAID 1: every write = 2 disk writes.
RAID 5 write penalty = 4 I/O ops; RAID 6 = 6 I/O ops.
Read Hit = data in cache (fast); Read Miss = fetch from disk.

⚠️ Common Mistakes to Avoid

Confusing track (a circle) with sector (a segment of a track).
Claiming SSDs are cheaper per GB than HDDs.
Saying RAID 0 provides redundancy — it does not.
Swapping Write Through and Write Back behavior.
Confusing SAN (block) with NAS (file).
Confusing scale-up (single box) with scale-out (cluster).
Mixing up where each type is implemented.
Confusing LUN masking (array-side access control) with zoning (switch-side).
Thinking concatenated metaLUN improves performance — it does not.
Saying thin provisioning reserves all space at creation.
Confusing front end (host side) with back end (disk side).
Thinking RAID 5 and RAID 6 have the same write penalty.

SSD vs HDD

Aspect	SSD	HDD
Moving parts	None	Yes (platters/heads)
Speed	Very fast	Slower
Latency	Low (microseconds)	Higher (milliseconds)
Cost per GB	Higher	Lower
Durability	More shock-resistant	Sensitive to shock

RAID Levels

Level	Technique	Min Disks	Fault Tolerance	Note
RAID 0	Striping	2	None	Fastest, no protection
RAID 1	Mirroring	2	1 disk	50% usable capacity
RAID 3	Striping + dedicated parity	3+	1 disk	Single parity disk; parallel I/O
RAID 5	Striping + distributed parity	3	1 disk	Single distributed parity
RAID 6	Striping + double parity	4	2 disks	Double parity
RAID 1+0	Mirror + stripe	4	1 per mirror	Speed + redundancy (RAID 10)

SAN vs NAS vs Object

Aspect	SAN (Block)	NAS (File)	Object
Access unit	Blocks	Files	Objects + metadata
Protocol	FC, iSCSI	NFS, SMB	HTTP/REST API
Performance	Highest	Moderate	Scalable, not low-latency
Best for	Databases, VMs	File sharing	Backups, media, cloud

Traditional vs Scale-out NAS

Aspect	Traditional (Scale-up)	Scale-out
Scaling	Upgrade single system	Add nodes to cluster
Architecture	Single file server	Cluster of nodes
Disruption	May require downtime	Non-disruptive expansion

Concatenated vs Striped MetaLUN

Aspect	Concatenated	Striped
Capacity	Yes	Yes
Performance	No gain	Improved
Expansion speed	Fast (no restripe)	Slow (restripe required)

Traditional vs Virtual (Thin) Provisioning

Aspect	Traditional (Thick)	Virtual (Thin)
Allocation	Upfront from RAID set	On demand from pool
Utilization	Lower (unused allocated)	Higher (over-provisioning)
Expansion	Create new LUN	Expand thin LUN + pool rebalance
Risk	Wasted space	Over-commit / out of space

Thin vs Thick Provisioning

Aspect	Thin	Thick
Allocation	On demand	Upfront (full)
Utilization	Higher	Lower
Risk	Over-commit / out of space	Wasted space
Performance	Slight overhead	Predictable

RAID Write Penalties

RAID Level	Writes per Host Write	I/O Operations
RAID 1	2 disk writes	2 writes
RAID 5	1 host write	4 I/O (2 read + 2 write)
RAID 6	1 host write	6 I/O (3 read + 3 write)

Week 4

FC SAN

Full lesson

📖 Must-Know Definitions

HBA: Host Bus Adapter connecting a server to the storage network.
Switched Fabric: FC topology using switches for dedicated, scalable connectivity.
E_Port: connects two FC switches via an Inter-Switch Link (ISL).
FC-2: framing, sequencing, and flow control layer.
WWPN: unique identifier for an individual FC port.
Frame: the smallest unit of data transfer in Fibre Channel.
FLOGI: Fabric Login, where a node registers with the fabric and receives an FCID.
RSCN: notification sent when fabric configuration changes.
Core-edge: storage connected to core tier switches; compute to edge.
BB_Credit: buffer-to-buffer flow control mechanism in Fibre Channel.
FC Frame: smallest unit of data transfer, containing SOF, header, payload, CRC, and EOF.

🎯 Key MCQ Points

HBA connects the host to the SAN.
Director = enterprise-class, high port count, highly available switch.
FC-AL = shared bandwidth, up to 126 nodes.
Switched fabric = scalable, dedicated bandwidth, most common today.
N = Node, F = Fabric (switch-to-node), E = Expansion (switch-to-switch / ISL).
FC-2 handles framing and flow control (BB_Credit).
FC-4 maps upper-layer protocols like SCSI.
FC-0 is the physical layer.
WWNN = node/adaptor; WWPN = port. Dual-port HBA = 1 WWNN + 2 WWPNs.
FC address assigned at fabric login; WWN is static (like MAC).
Name Server maps WWNs to dynamic FC addresses.
Largest → smallest: Exchange → Sequence → Frame.
Order: FLOGI → PLOGI → PRLI.
FLOGI is with the fabric; PLOGI is port-to-port.
FLOGI uses Fabric Login Server at address FFFFFE.
Fabric Login Server = FFFFFE; Name Server = FFFFFC; Fabric Controller = FFFFFD.
Without zoning, RSCNs go to ALL nodes in the fabric.
Link aggregation = multiple ISLs → one logical ISL with combined bandwidth.
BB_Credit = FC flow control (buffer-to-buffer).
WWN zoning = flexible/soft; Port zoning = secure/hard.
Zoning is on the SWITCH; LUN masking is on the ARRAY.
Frame = 5 parts: SOF, header, payload, CRC, EOF.
Frame is the SMALLEST FC data unit.

⚠️ Common Mistakes to Avoid

Confusing a hub (shared bandwidth) with a switch (dedicated paths).
Thinking FC-AL gives dedicated bandwidth — it's shared.
Mixing N_Port (node) with F_Port (switch side).
Reversing the order — FC-0 is physical, FC-4 is protocol mapping.
Confusing WWNN (node-level) with WWPN (port-level).
Reversing the hierarchy order.
Mixing FLOGI (fabric) with PLOGI (port-to-port).
Confusing Name Server (FFFFFC) with Fabric Login Server (FFFFFE).
Confusing mesh topology with arbitrated loop (FC-AL).
Confusing zoning (switch-side) with LUN masking (array-side).
Confusing frame (smallest) with exchange (largest).

FC Topologies

Topology	Bandwidth	Scalability	Note
Point-to-Point	Dedicated	2 devices only	Direct link
FC-AL	Shared	Up to 126 devices	Loop, legacy
Switched Fabric	Dedicated	Very high	Modern standard

FC Layers

Layer	Role
FC-4	Protocol mapping (e.g. SCSI, IP over FC)
FC-3	Common services
FC-2	Framing, sequencing, flow control
FC-1	Encoding / decoding
FC-0	Physical layer (media, cables, speed)

FC Data Hierarchy

Unit	Contains	Note
Exchange	Multiple sequences	Largest unit (a full operation)
Sequence	Multiple frames	Set of related frames
Frame	Payload	Smallest transmission unit

Fabric Design Patterns

Topology	Description	Note
Single-switch	One switch, all nodes connected	No ISLs needed
Full mesh	Every switch connected to every other	Max 1 ISL hop
Partial mesh	Not all switches interconnected	Some paths longer
Core-edge	Edge + core tiers; storage on core	Scalable enterprise design

WWN Zoning vs Port Zoning

Aspect	WWN Zoning	Port Zoning
Based on	Device WWN	Switch port
Flexibility	High (move cables freely)	Lower (tied to port)
Security	Lower (WWN spoofing)	Higher
Also called	Soft zoning	Hard zoning

Week 5

IP SAN

Full lesson

📖 Must-Know Definitions

IP SAN: block-level SAN transport using IP-based protocols.
iSCSI: protocol that carries SCSI commands over IP networks.
TOE: TCP Offload Engine that processes TCP in hardware.
Bridged iSCSI: uses a bridge/gateway between iSCSI and FC.
iSNS: Internet Storage Name Service for automated iSCSI discovery.
FCIP: encapsulates FC frames in IP to link distant SANs.
IQN: unique iSCSI name identifying an initiator or target.
Dual-protocol storage: array with both FC and iSCSI front-end ports.
FCIP encapsulation: wrapping FC frames in IP/TCP for SAN extension.

🎯 Key MCQ Points

IP SAN protocols: iSCSI (host-to-storage) and FCIP (SAN-to-SAN extension).
Leveraging IP reduces cost vs deploying new FC infrastructure.
Initiator = host (sends); Target = storage (serves).
iSCSI = SCSI over TCP/IP.
iSCSI HBA = full hardware offload, lowest CPU load.
TOE NIC offloads only the TCP part.
Bridged iSCSI needs a gateway to reach FC storage.
iSNS = automatic discovery (DNS-like); SendTargets = manual.
iSNS discovery domains function like FC zones.
FCIP = tunnel FC over IP (SAN extension), NOT host-to-storage like iSCSI.
EX_Port = no fabric merge; VE_Port = fabric merge via virtual ISL.
iSCSI is a SESSION-layer protocol (Layer 5).
iSCSI address = IP + TCP port + IQN name.
Stack order: SCSI → iSCSI → TCP → IP → Ethernet.
Dual-protocol = FC + iSCSI on same array, no bridge.
Bridged iSCSI is only needed when storage is FC-only behind a gateway.
FCIP encapsulation: FC frame → FCIP → TCP → IP.
FCP = SCSI over FC at FC-4 layer.

⚠️ Common Mistakes to Avoid

Thinking IP SAN only means iSCSI — FCIP is also an IP SAN protocol.
Swapping initiator (client) and target (storage).
Thinking a software initiator offloads CPU — it uses the most CPU.
Thinking native iSCSI requires a bridge.
Confusing SendTargets (manual) with iSNS (automatic).
Confusing FCIP (SAN-to-SAN extension) with iSCSI (host-to-storage).
Placing iSCSI at the transport layer — it's session layer.
Assuming iSCSI always needs a bridge — native or dual-protocol arrays do not.
Confusing FCP (SCSI over FC) with FCIP (FC tunneling over IP).

iSCSI Connection Types

Type	Processing	CPU Load	Cost
NIC + Software	CPU does TCP + iSCSI	High	Lowest
TOE NIC	Hardware does TCP	Medium	Medium
iSCSI HBA	Hardware does TCP + iSCSI	Low	Highest

Native vs Bridged iSCSI

Aspect	Native	Bridged
Target	iSCSI-capable storage	FC storage via gateway
Bridge needed?	No	Yes
Use case	New iSCSI deployments	Reusing FC investment

iSCSI Protocol Stack

Layer	Protocol	Function
Application	SCSI	Storage commands and data
Session (L5)	iSCSI	Login, auth, discovery, session
Transport (L4)	TCP	Reliable delivery
Network (L3)	IP	Routing
Data Link (L2)	Ethernet	Frames on the wire

Week 6

FCoE

Full lesson

📖 Must-Know Definitions

CNA: Converged Network Adapter combining NIC and HBA functions.
VN_Port: virtual node port on a CNA in FCoE.
VSAN: virtual SAN logically partitioning a Fibre Channel fabric.
FCoE with existing FC SAN: FCoE switches interconnect CEE to FC storage.
PFC: Priority Flow Control providing lossless behavior per traffic class.
FPMA: Fabric Provided MAC Address derived from FC-MAP and FCID.
FC-MAP: 24-bit fabric-assigned prefix used in FPMA addressing.

🎯 Key MCQ Points

CNA = NIC + HBA combined (converges LAN + SAN on one adapter).
FCF = FCoE Forwarder does the FC switching role.
FCoE needs LOSSLESS Ethernet (no TCP/IP, unlike iSCSI/FCIP).
V = virtual: VN→N, VF→F, VE→E port roles over Ethernet.
VLAN → Ethernet; VSAN → FC SAN. FCoE switch maps VSAN to dedicated VLAN.
End-to-end FCoE = both compute and storage are FCoE-native.
PFC = lossless via per-priority pause (key for FCoE).
ETS = bandwidth allocation per traffic class.
DCBX = negotiates DCB parameters between devices.
FIP = FCoE Initialization Protocol (discovery + login).
FPMA = fabric provides MAC (FC-MAP + FCID); SPMA = server provides MAC.
FCoE reduces adapters, cables, power, and space.
FPMA MAC = FC-MAP (24-bit) + FC address/FCID (24-bit).
FCoE uses lossless Ethernet — NOT TCP/IP.

⚠️ Common Mistakes to Avoid

Thinking FCoE runs over TCP/IP — it runs directly on lossless Ethernet.
Forgetting the 'V' indicates the virtualized FCoE version of FC ports.
Mixing VLAN (LAN) and VSAN (SAN) domains.
Using VSAN VLANs for regular LAN traffic.
Thinking FCoE always replaces FC storage — it can bridge to existing FC SAN.
Confusing PFC (pause/lossless) with ETS (bandwidth allocation).
Swapping FPMA (fabric-provided) and SPMA (server-provided).
Running FCoE over standard lossy Ethernet without DCB/PFC.

FCoE Virtual Ports vs FC Equivalents

FCoE Port	FC Equivalent	Role
VN_Port	N_Port	Node port (on CNA)
VF_Port	F_Port	Fabric port (on FCF)
VE_Port	E_Port	Switch-to-switch (FCF-to-FCF)

VLAN vs VSAN

Aspect	VLAN	VSAN
Network	Ethernet / LAN	Fibre Channel / SAN
Purpose	Segment broadcast domains	Segment fabric / isolate SAN traffic
Isolation	Layer 2 LAN	FC fabric services

FCoE Connectivity Modes

Mode	Compute	Storage	Use Case
With existing FC SAN	CNA (FCoE)	FC ports	Leverage existing FC investment
End-to-end FCoE	CNA (FCoE)	FCoE ports	New greenfield deployment

FCoE vs iSCSI vs FCIP

Protocol	Network	Use Case	TCP?
FCoE	Lossless Ethernet	Converged LAN+SAN in DC	No
iSCSI	Standard IP/Ethernet	Host to block storage	Yes
FCIP	IP/WAN	Remote SAN extension	Yes

Week 10

Cloud Networking

Full lesson

📖 Must-Know Definitions

Scatter-Gather: one request distributed to many servers, responses aggregated.
Flow size: data volume of a single flow (mice = small, elephant = large).
TCP Incast: throughput collapse caused by synchronized many-to-one traffic.
ToR: Top-of-Rack switching with a switch in every rack.
Fabric Extender: extends switch fabric into racks as remote line cards.
East-West traffic: server-to-server communication within the data center.
Clos network: multi-stage spine-leaf fabric with high bisection bandwidth.

🎯 Key MCQ Points

Scatter-gather causes synchronized bursts → leads to TCP Incast.
~76% of cloud DC traffic is East-West (server-to-server within DC).
Know all five: volume, locality, concurrent flows, arrival rate, flow size.
Most traffic is cluster-local (~57.5%).
Mice = small flows; Elephant = large flows.
TCP Incast = many-to-one synchronized bursts → buffer overflow → throughput collapse.
ToR = switch per rack (short cables, more switches).
EoR = switch per row (longer cables, fewer/centralized switches).
Why ToR: east-west traffic, copper in-rack, modular per-rack, unified fabric ready.
Why not ToR: more switches, power, STP instances, port utilization.
Why EoR: fewer switches, fewer STP instances, Layer 1 rack connections.
Why not EoR: expensive bulky copper, cable management challenges.
FEX = remote line card; fewer switches to manage + in-rack copper.
Combines ToR cabling benefits with EoR centralized management.
76% of cloud DC traffic is East-West.
Virtualization increases East-West traffic.
DC traffic grows ~25% per year (Cisco Global Cloud Index).
Clos / spine-leaf = many equal paths → great for East-West traffic.
East-West = inside DC; North-South = in/out of DC.
Tree networks bottleneck at the core.

⚠️ Common Mistakes to Avoid

Not linking scatter-gather to TCP Incast congestion.
Forgetting locality as a traffic characteristic.
Describing TCP Incast as one-to-many — it's many-to-one.
Swapping ToR and EoR cabling/switch-count tradeoffs.
Treating FEX as a fully independent switch — it's managed by a parent.
Underestimating East-West traffic — it dominates cloud DCs.
Calling East-West traffic 'in and out of the data center' — that's North-South.

Traffic Locality (Typical %)

Scope	Percentage
Rack	12.9%
Cluster	57.5%
Data Center	11.9%
Inter-DC	17.7%

ToR vs EoR

Aspect	ToR (Top-of-Rack)	EoR (End-of-Row)
Switch location	In each rack	At end of the row
Cabling	Short, within rack	Longer, across row
Switch count	More switches	Fewer switches
Management	More devices to manage	Centralized

Tree vs Clos Network

Aspect	Tree	Clos (Spine-Leaf)
Paths	Single/limited	Many equal-cost paths
Bottleneck	At aggregation/core	Avoided (scale-out)
Cost	Expensive large switches	Many cheap small switches
Scalability	Poor (N² problem)	High bisection bandwidth
Best for	Traditional North-South	East-West heavy workloads

East-West vs North-South Traffic

Direction	Meaning	Example
East-West	Server-to-server inside the DC	VM-to-VM, distributed compute
North-South	In/out of the data center	Client request from the internet

Week 12

Control Layer

Full lesson

📖 Must-Know Definitions

Unified Manager: centralized control across all resource types.
Resource discovery: identifying available compute, network, and storage.
Relative allocation: proportional resource sharing using weights/shares.
Memory page sharing: deduplicating identical memory pages across VMs.
Tiering: automatically relocating data across storage tiers by access frequency.
Multipathing: using multiple paths between host and storage for redundancy and load balancing.
Service catalog: predefined offerings for cloud resource requests.
Element manager: software managing a single infrastructure component type.

🎯 Key MCQ Points

Element manager = single domain; Unified manager = centralized across all.
Discovery is scheduled periodically or triggered on infrastructure change.
Network discovery includes VLAN IDs, VSAN IDs, and zones.
Gold > Silver > Bronze in quality.
Relative = shares/proportional (Platinum 2X vs Gold 1X).
Absolute: VM won't power on if minimum resources unavailable.
Hyperthreading = 1 core → 2 logical CPUs.
Memory page sharing = deduplication of identical pages.
Tiering = move data between tiers by 'temperature' (hot/cold).
Link aggregation = more bandwidth + redundancy by bonding links.
Multipathing = multiple paths to STORAGE (redundancy/load balancing).
NIC teaming = grouping NICs for failover/throughput.
Workflow order: Discovery → Grading → Allocation → Provisioning.
Service catalog defines what consumers can request.
Control layer receives service-layer requests (e.g. VM with 4 GB RAM).
Element manager tasks: config, expand, troubleshoot, monitor.
Storage element manager: RAID, LUN masking. Network: VLANs, zoning.

⚠️ Common Mistakes to Avoid

Confusing element (single type) with unified (all types) manager.
Forgetting that discovery precedes management/allocation.
Mixing relative (proportional) with absolute (fixed) allocation.
Thinking hyperthreading adds physical cores — it adds logical ones.
Confusing tiering (data placement) with cache tiering (SSD cache layer).
Confusing NIC teaming (NICs/LAN) with multipathing (paths to storage).
Skipping discovery before provisioning — resources must be inventoried first.
Expecting element manager to manage all resource types — that's unified manager.

What Discovery Collects

Domain	Discovered Items
Compute	Blade servers, CPU speed, memory, VM-to-physical mapping
Network	Switch model, adapters, VLAN/VSAN, QoS, topology, zones
Storage	System type, drive type, capacity, RAID, pools, mappings

Relative vs Absolute Allocation

Aspect	Relative	Absolute
Basis	Shares / weights	Fixed amount
Behavior	Proportional under contention	Guaranteed reservation/limit
Example	Platinum 2X vs Gold 1X shares	2–4 GB RAM bounds; VM won't start if min unavailable

Storage Grading Examples

Tier	Drives	Tiering	RAID
Gold	Flash + FC + SATA	Yes	RAID 5
Silver	Flash + FC + SATA	Yes	RAID 1+0
Bronze	FC only	No	RAID 5