Technical Specification

Table of Contents

1. Project Overview
2. Infrastructure Specification
   1. Compute
   2. Networking
   3. Storage
3. Region Coverage
4. Protocol Specifications
   1. WebSocket Echo
   2. HTTP Ping
   3. Blob Storage
5. Naming Conventions
6. Deployment Scripts
7. Frontend Specification
8. Performance Baseline
9. Key Findings
10. Constraints & Limitations

---

Project Overview

Attribute	Value
Name	Azure APAC Latency Test
Purpose	Measure and compare network latency across Azure APAC regions
Protocols	WebSocket, HTTP, Azure Blob Storage
Regions	14 Azure APAC regions
Client	Angular 17 SPA (browser-based measurement)
Infrastructure	Azure CLI deployment (no IaC frameworks)

---

Infrastructure Specification

Compute

Property	Value
VM SKU	Standard_B2s (2 vCPU, 4 GB RAM)
OS Image	Canonical:ubuntu-24_04-lts:server:latest
Provisioning	cloud-init (automatic on first boot)
Admin User	azureuser (SSH key auth only)
Services	nginx (port 80), ws-echo (port 8080)
Count	14 (one per region)

Networking

Property	Value
Public IP	Static SKU Standard, one per VM
NSG Rules	Inbound TCP 80 (HTTP), TCP 8080 (WebSocket)
DNS	None (direct IP access)
Load Balancer	None (single VM per region)

Storage

Property	Value
Account Type	StorageV2, Standard_LRS
Feature	Static Website hosting enabled
Payload	latency-test.txt (1 KB)
CORS	All origins allowed (for browser access)
Container	$web (static website root)
Count	14 (one per region)

---

Region Coverage

#	Region ID	Display Name	City
1	australiacentral	Australia Central	Canberra
2	australiaeast	Australia East	Sydney
3	australiasoutheast	Australia Southeast	Melbourne
4	newzealandnorth	New Zealand North	Auckland
5	eastasia	East Asia	Hong Kong
6	southeastasia	Southeast Asia	Singapore
7	japaneast	Japan East	Tokyo
8	japanwest	Japan West	Osaka
9	koreacentral	Korea Central	Seoul
10	koreasouth	Korea South	Busan
11	centralindia	Central India	Pune
12	southindia	South India	Chennai
13	indonesiacentral	Indonesia Central	Jakarta
14	malaysiawest	Malaysia West	Kuala Lumpur

---

Protocol Specifications

WebSocket Echo

Property	Value
Port	8080
Path	/
Protocol	RFC 6455 (WebSocket)
Server	Node.js + ws library
Behavior	Echo exact message back

Message Format (Client → Server → Client):

{ "t": 1716710400000 }

Measurement Protocol:

1. Client opens WebSocket connection to ws://<ip>:8080/
2. On open, client sends 3 sequential probe messages
3. Each probe: RTT = Date.now() - JSON.parse(response).t
4. Wait for echo before sending next probe
5. Reported latency = average of 3 RTTs
6. Client closes connection

Error Handling:

Scenario	Behavior
Connection timeout (>5s)	Status: timeout, latency: null
Connection refused	Status: error, latency: null
Echo timeout (>3s)	Status: timeout for that probe
WebSocket error event	Status: error, latency: null

---

HTTP Ping

Property	Value
Port	80
Method	HEAD
Path	/
Server	nginx
Response	200 OK (no body)

Measurement Protocol:

1. Send HEAD http://<ip>/ request
2. RTT = responseTime - requestTime
3. Repeat 3 times, report average
4. Each request includes TCP connection setup

Notes:

• No TLS (plain HTTP) to isolate network latency
• HEAD method minimizes response payload
• Includes TCP handshake overhead on each request

---

Blob Storage

Property	Value
Endpoint	https://<account>.z23.web.core.windows.net/latency-test.txt
Method	HEAD
Size	1 KB payload
Auth	Anonymous (public static website)

Measurement Protocol:

1. Send HEAD request to static website endpoint
2. RTT = responseTime - requestTime
3. Repeat 3 times, report average

Notes:

• Includes DNS resolution + TLS handshake + storage front-end processing
• Higher baseline latency than WebSocket/HTTP (~300-400ms overhead)
• Useful for measuring Azure Storage infrastructure performance

---

Naming Conventions

Resource	Pattern	Example
Resource Group	rg-latency-{region}	rg-latency-southeastasia
VM	vm-latency-{region}	vm-latency-japaneast
Public IP	pip-latency-{region}	pip-latency-koreacentral
NSG	nsg-latency-{region}	nsg-latency-australiaeast
NIC	nic-latency-{region}	nic-latency-centralindia
Storage Account	latency{region}	latencysoutheastasia

---

Deployment Scripts

Script	Purpose	Idempotent
deploy/ensure-vms-ready.sh	Create/start VMs + apply NSG rules	✅
deploy/create-storage-accounts.sh	Provision storage with static website	✅
deploy/upload-blob-payload.sh	Upload 1KB test file to all accounts	✅
deploy/configure-blob-cors.sh	Enable CORS for browser access	✅
deploy/create-resource-groups.sh	Create RGs in all 14 regions	✅
deploy/teardown.sh	Delete all resource groups	❌

---

Frontend Specification

Property	Value
Framework	Angular 17
Charts	Chart.js
Styling	Tailwind CSS
Build	ng build --configuration production
Hosting	VM (Singapore) or ng serve locally

Features:

• Real-time latency measurement from browser
• Visual comparison across all 14 regions
• Protocol selection (WebSocket / HTTP / Blob)
• Auto-refresh with configurable interval
• Results export

---

Performance Baseline

Reference measurements from Singapore (Southeast Asia):

Region	WebSocket RTT	Distance Factor
Indonesia Central	20 ms	Closest
Southeast Asia	36 ms	Local
Malaysia West	39 ms	Adjacent
East Asia	69 ms	Regional
South India	70 ms	Regional
Central India	89 ms	Medium
Korea South	93 ms	Medium
Japan West	103 ms	Far
Korea Central	104 ms	Far
Japan East	111 ms	Far
Australia Southeast	124 ms	Far
Australia East	135 ms	Distant
Australia Central	137 ms	Distant
New Zealand North	158 ms	Farthest

---

Key Findings

• B1s → B2s upgrade reduced latency 60-90% (CPU throttling inflates RTT on B1s)
• Blob latency adds ~300-400ms overhead vs WebSocket (DNS + TLS + storage front-end)
• New regions (Indonesia, Malaysia) show fastest network RTT but less mature storage
• Accelerated Networking unavailable on B-series; requires D/F-series for 1-3ms improvement
• Physical distance is the dominant latency factor (speed of light in fiber ≈ 5μs/km)

---

Constraints & Limitations

Constraint	Reason
No TLS on WebSocket/HTTP	Isolates network latency from TLS overhead
No load balancer	Single VM per region keeps cost low
B2s only	Avoids CPU throttling; D-series costs 4x more
No VNet peering	Tests public internet path (user perspective)
Browser-only measurement	No server-to-server or Azure backbone testing
14 regions	APAC focus; excludes Americas/EMEA

---

← Home Architecture →