Transcript Document

Migration of Real Product
into Windows Azure
Lessons Learned
Agenda
1.
2.
3.
4.
5.
Solution Architecture Concept
Think as Azure Developer
Minimize Latency
Master Windows Azure SQL Database
Entity Framework with Windows Azure SQL
Database
6. Cloud Services - Tips and Tricks
7. Best approaches for Windows Azure Diagnostics,
Logging and Monitoring
8. Summary
Customer 1
Desktop
Clients
Data
Layer
Application
Layer
Integration
Layer
Client Layer
(on-premise)
Solution Architecture Concept
Embedded
Clients
Traffic
Manager
Reporting
Storage / SQL Azure
Compute
Customer N
Connect
Access
Control
Messaging
Cloud Service 1
Database
Desktop
Clients
Tables
Cloud Service N
Caching
Embedded
Clients
Data Sync
Needs in Technologies / Components
Think as Azure Developer!
The three most
important areas to
master in order to
design your
application to run
at the appropriate
scale, performance,
and for the best
costs
Think as Azure Developer – Minimize Latency
Response Time = 2 x (Internet Latency + DC Latency) +
App Logic Exec Time + DB Query Exec Time
Web/Worker Roles
Storage / SQL Azure
Think as Azure Developer – Minimize Latency
Response Time = 2 x (Internet Latency + DC Latency) +
App Logic Exec Time + DB Query Exec Time
Think as Azure Developer – Minimize Latency
• To Minimize Internet Latency
 Select a data center closest to majority of your
users
 Use CDN
 Combine round-trips from the client application
to reduce any physical network overhead
associated between tiers
• To Minimize Data Center Latency
 Use Cache where applicable
 Combine round-trips from services to Data
Storage
Master Windows Azure SQL Database
• Scale it out
 Use several separate Azure SQL Databases
 Use Federations
 Consider using NoSQL storage for plain data
 Use multiple subscriptions
• Minimize number of round-trips
 Use local or Azure Cache when possible
 Replace Ad-hoc queries with Stored Procedures
 Batch up queries
 Use Bulk loading instead of individual inserts
Master Windows Azure SQL Database
• Expect some operations need to be retried
(e.g. due to transient faults, throttling, etc.)
 Avoid long-running queries / split operations
into smaller chunks
 Make your operations easily restartable
• Determine how much you can stress test
your database before getting throttled
 Reduce number of application level connections
 Scale out data model adding more databases
Master Windows Azure SQL Database
• Follow traditional best-practices for Data
Access Layer
 Retrieve only data that you need
 Keep transactions as short as possible
 Open the connection late and close it ASAP
• Examine execution plan for each “slow”
query, and search areas for improvement
• Examine Database object schemas and
constraints
Entity Framework with Windows Azure
SQL Database
Lazy Loading
using (var context = new AdventureWorksDWEntities())
{
var query = contect.Customers.Where(c =>
c.InternetSales.Any()).Take(100);
var customers = query.ToList();
EnumerateCustomers();
}
Eager Loading
using (var context = new AdventureWorksDWEntities())
{
var query =
contect.Customers.Include(“InternetSales”).Where(c
=> c.InternetSales.Any()).Take(100);
var customers = query.ToList();
EnumerateCustomers();
}
Entity Framework with Windows Azure
SQL Database
Figure 1
Comparing Eager Loading to Lazy
Loading from a Local Database
Figure 2
Comparing Eager Loading to Lazy
Loading from SQL Azure
Cloud Services – Best Practices
• Explicit design to scale-out, NOT scale-up
• Do not store state on individual instances (“make
everything stateless”)
• Do not assume subsequent requests will hit the same
instance
• Prefer asynchronous calls
• Design for availability
• Avoid single point of component failure (multiple
web/worker instances, spread across multiple
fault/upgrade domains)
• Fall back to multiple components to mitigate the impact
of a specific service being offline / unavailable
Cloud Services – Best Practices
• Handle failure gracefully
 For transient failures, provide appropriate retry
mechanisms to reconnect or resubmit work
 For other failure events, provide rich instrumentation on
the failure event and a suitable error message back to
the user.
 Fall back to a different service or workflow where
possible
• Isolate services and client APIs behind
generic interfaces allow for replacement or
side-by-side evaluation
Cloud Services – Optimize Communication
• Use REST-full services if possible
• Simpler code paths and easier expose data
• Use more efficient JSON or binary serialization
frameworks
• Serialization/Deserialization hugely impactful on throughput
and latency
• Use network friendly data transfer objects
(DTOs)
• Do NOT mix data/state and logic in DTOs
• Avoid dependencies (if you cannot easily serialize a DTO class,
you have a bad design)
• Use asynchronous communication
• Minimize blocking IIS dispatch pipeline
Classic 1:1 Migration into Azure
• What is wrong
with this?
Web Roles
SQL Azure
“Smart” Migration into Azure
• Scale storage
• Decouple
receiving work
from executing
work
• Scale work
tasks via
queues
Web Roles
Worker Roles
SQL Azure / Storage
Windows Azure Telemetry & Diagnostics
• Collect
 Tune infrastructure performance counters and system
events
 Instrument application logging
• Analyze
 Grab real-time and historical event logs
 Aggregate/Filter required data
• Consume
 Make well-considered correction actions
 Publish analytics in a consumable format
Windows Azure Diagnostics – Internal View
Windows Azure Diagnostic Tips
• Do not rely on Default Diagnostic Monitor
configuration
• Remove unused and add required performance
counters / logs
• Apply appropriate log level filters
• Set appropriate scheduled transfer periods (5 – 15
minutes)
• Use dedicated storage account for
diagnostics data
Windows Azure Logging Tips
• Optimize custom application logging
 Log all API calls with context, method, unique
identifiers of operation, timestamps (latency)
and result
 Use common logging libraries (e.g. log4net,
Nlog)
 Use Custom Data Source in Diagnostic Monitor
to copy files periodically to blob storage
Summary
• Be a pro Azure Developer:
 Think about performance and latency!
 Think about operational costs!
 Think about proper logging and monitoring of
your solution!
Questions Time
Thank you for your time!
Questions?
Заповни Анкету
Виграй Приз
http://anketa.msswit.in.ua
Migration of Real
Product into Windows
Azure
Lessons Learned
By
Oleksii Syrotenko
Development Team Leader, EPAM Systems
E-mail: [email protected]
Skype: oleksii.syrotenko