Scaling asp.net websites to millions of users

Scaling ASP.NET websites
from thousands to millions
of users
Lessons learnt from scaling Web 2.0 apps to millions of
users

Omar AL Zabir
Chief Architect, SaaS Platform, BT
omaralzabir.com
omaralzabir@gmail.com
Twitter.com/omaralzabir

Book “Building Web 2.0 portal using ASP.NET 3.5” from O‟Reilly

How to blow up a website

 Get some super fast fiber broadband connection.
 Run this on couple of quad core desktops:

for (int j = 0; j < 100; j ++)
{
for (int i = 0; i < 100; i ++)
{
var client = new WebClient();
client.DownloadStringAsync(
new Uri("http://www.microsoft.com/MISReport.aspx");
}
Thread.Sleep(500);
}

Prevent App Level DOS attack

 Application Level DOS attacks are attempts to hit
your expensive pages too frequently so that you
have 100% CPU and your site goes down. For ex,
hitting some expensive Report page continuously.
 You can run out of ASP.NET Threads and stop
responding completely to any request.
 I‟m not talking about Network level DOS attacks
like TCP SYN flood or DDOS attacks that hardware
firewalls can prevent.
 Firewalls don‟t block legitimate requests.

Prevent App level DOS attacks
 Protect only expensive pages. Pages that are unlikely to
get hit too many times within a short duration.
 Build a HttpModule and hook on OnInit event.
 Store which IP is making how many number of hits in
last 10 minutes. Store the table in some in-memory
cache.
 If the threshold has exceeded, stop responding to that
IP. Call Response.End()
 Solution is here:
http://tinyurl.com/omarDOS

IP Hits in last 10 mins
128.0.0.1 500
128.0.0.2 345

ASP.NET ProcessModel Optimization
 ASP.NET ProcessModel Defaults
 maxWorkerThreads = 20
 maxIOThreads = 20
 memoryLimit = 60
 Nowadays, servers are way too
powerful. You don‟t need to be
conservative.
 Change default process model
setting in machine.config to make
best use of CPU power.

ASP.NET Pipeline Optimization

 Default ASP.NET Pipeline has several components that intercept
each and every request. Thus they add extra processing overhead
on every request.
 Multiply the overhead by hundreds of requests per second – you
get a significant overhead.

ASP.NET Pipeline Optimization
 Remove what you don‟t need.
 If you are using Forms Authentication, SQL Server
storage, no web.config based role based permission then
you can remove most of them:

Prevent large ASP.NET cookies on
static content
 Each and every request, even static files, get the
ASP.NET cookies sent.
 517 bytes of worthless data per request.
 Avg pages have 40 resources. 40 x 517 = 20 KB.
 1M page view = 20 GB of data upload to server.

Cookie:
.DBANON=w3kYczsH8Wvzs6MgryS4JYEF0N-8ZR6aLRSTU9KwVaGa
ydD6WwUHD7X9tN8vBgjgzKf3r3SJHusTYFjU85yYfnunyCeuExcZs8
95JK9Fk1HS68ksGwm3QpxnRZvpDBAfJKEUKee2OTlND0gi43qwwtI
PLeY1; ASP.NET_SessionId=bmnbp155wilotk45gjhitoqg; .DBAUTH
12=2A848A8C200CB0E8E05C6EBA8059A0DBA228FC5F6EDD2940
1C249D237812344C15B3C5C57D6B776037FAA8F14017880E57B
DC14A7963C58B0A0B30229AF0123A6DF56601D814E75525E7DC
A9AD4A0EF200832B39A1F35A5111092F0805B0A8CD3D2FD5E3A
B6176893D86AFBEB68F7EA42BE61E89537DEAA3279F3B576D0C
44BA00B9FA1D9DD3EE985F37B0A5A134ADC0EA9C548D

Prevent ASP.NET cookies on static
content
 Setup a new website in IIS, map to the same code folder.
 Map static.yoursite.com host header to that website.
 Prefix all css, js, image links with http://static.yoursite.com/
 Add a Global.asax‟s EndResponse event on the new website.
HttpContext context = HttpContext.Current;
if (context.Request.Url.Host == “static.yoursite.com”)
{
List<string> cookiesToClear = new List<string>();
foreach (string cookieName in context.Request.Cookies)
{
HttpCookie cookie =
context.Request.Cookies[cookieName];
cookiesToClear.Add(cookie.Name);
}
foreach (string name in cookiesToClear)
{
HttpCookie cookie = new HttpCookie(name,
string.Empty);
cookie.Expires = DateTime.Today.AddYears(-1);
context.Response.Cookies.Set(cookie);

System.net optimization

 If you are using HttpWebRequest or WebClient or any
other TCP/IP operation, increase the max connection
limit.
 Default is 2 per IP.
 WCF service calls are limited by this setting.
 Unless you suspect rogue clients, set some reasonably
high number on webservers, but moderate number on
desktop clients.

System.net default setting is
suboptimal

Bottleneck

Max 2 concurrent calls

ASP.NET Profile Provider

 Anonymous Provider creates one anonymous user in
database on every first hit to Profile object‟s Custom
properties:

 The SP aspnet_Profile_GetProperties gets called when
you access Profile object for first time in a request. And
this SP is slooooooooow!

 The slow SP that gets fired when you access custom Profile
properties:

CREATE PROCEDURE [dbo].[ aspnet_Profile_GetProperties]
@ApplicationName nvarchar(256), @UserId …
AS
BEGIN

DECLARE @ApplicationId uniqueidentifier
SELECT @ApplicationId = NULL
SELECT @ApplicationId = ApplicationId
FROM dbo.aspnet_Applications
WHERE LOWER(@ApplicationName) = LoweredApplicationName

…
…
…
IF (@@ROWCOUNT > 0)
BEGIN
UPDATE dbo.aspnet_Users
SET LastActivityDate=@CurrentTimeUtc
WHERE UserId = @UserId
END
END


 The slow SP‟s execution plan

 Don‟t update LastActivityDate when Profile object is
loaded. Do it only when Profile object is updated.
 Update once every hour or so. If LastActivityDate <
DateTime.Now.AddHours(-1) then update. No need to do
per request or too frequently.
 Hard code the Application ID to avoid one lookup inside
the SP.
 tinyurl.com/omarGetProp

ASP.NET Membership
 ASP.NET Membership Provider Stored Procs use default
transaction isolation level, which is Serializable.

Transaction (Process ID ##) was deadlocked on lock resources
with another process and has been chosen as the deadlock
victim. Rerun the transaction.
Timeout Expired. The Timeout Period Elapsed Prior To
Completion Of The Operation Or The Server Is Not Responding.

ALTER PROCEDURE [dbo].[aspnet_Profile_GetProperties]
@ApplicationName nvarchar(256),
@UserName nvarchar(256),
@CurrentTimeUtc datetime
AS
BEGIN

SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED;

ASP.NET Membership Query
 Common queries that brings down hell:
 Select * from aspnet_users where UserName = „blabla‟
 Select * from aspnet_membership where Email =
“someone@somewhere.com”
 What‟s wrong with these queries?

ASP.NET Membership Queries

 Let's look at the indexes:
 Table: aspnet_users
 Clustered Index = ApplicationID, LoweredUserName
 NonClustered Index = ApplicationID, LastActivityDate
 Primary Key = UserID
 Table: aspnet_membership
 Clustered Index = ApplicationID, LoweredEmail
 NonClustered = UserID
 Table: aspnet_Profile
 Clustered Index = UserID

 DO NOT use Email or UserName fields in WHERE clause. They are not
part of the index instead LoweredUserName and LoweredEmail fields
are in conjunction with ApplicationID field. All queries must have
ApplicationID in the WHERE clause.

A love story

 .NET 3.0 was released, woohoo!
WCF!
Lambda Expressions!!
Linq to SQL!!!
 Upgraded to .NET 3.0, top to bottom.
 Major deployment over the weekend.
 Monday 9 AM, peak traffic.
 No response from site.

Linq to SQL is not suitable for high
volume web applications
 Linq to SQL is not optimized for web application.
 No disconnected entity support.
 Entities are not serializable.
 Linq to sql expressions consume high CPU when
compiled to SQL.

var query = from widget in dc.Widgets
where widget.ID == id && widget.PageID == pageId
select widget;

var widget = query.SingleOrDefault();

How bad Linq to SQL is?

Source: JD Conley‟s blog

Fixing Linq to SQL
 Convert all queries to Compiled Queries.
 tinyurl.com/omarLINQ

Linq to SQL transaction deadlocks
 Large table, high read and medium write, causes query
timeouts, high locks, transaction deadlock because of
SERIALIZATION isolation level.

Transaction (Process ID ##) was deadlocked on lock resources
with another process and has been chosen as the deadlock
victim. Rerun the transaction.
Timeout Expired. The Timeout Period Elapsed Prior To
Completion Of The Operation Or The Server Is Not Responding.
using (var db = new YourDataContext2())
{
db.Connection.Open();
db.ExecuteCommand("SET TRANSACTION ISOLATION LEVEL READ
UNCOMMITTED;");

var user = db.aspnet_Users.First();
var pages = user.Pages.ToList();
}

Linq to SQL transaction deadlocks
 Connection does not close!
 Bug in Data Context code.
 Solution is to override the Dispose.

class YourDataContext2 : YourDataContext, IDisposable {
public new void Dispose() {
if (base.Connection != null)
if (base.Connection.State != System.Data.ConnectionState.Closed) {
base.Connection.Close();
base.Connection.Dispose();
}
base.Dispose();
}
}

Content Delivery Network (CDN)
 CDN cache and deliver content from their servers that
are closest to users‟ computers. The closer servers are
to end user, the shorter roundtrip time for every request.
For ex, Akamai has servers almost every city in the
world.
 Content from CDN nodes get served faster with lower
latency than coming from your servers.
 CDN Nodes have better caching and compression
algorithms.
 CDN nodes can offload your server and network from
delivering static files. Thus better throughput for
dynamic content.

Content Delivery Network

0.7
sec

Washington, DC

2.2
sec

Australia

Without CDN

www.yoursite.com

How CDN works

static.yoursite.com

www.yoursite.com

Two types of CDN

 Static – you upload the files to CDN and
they give you an URL. E.g.
yoursite.cachefly.net/logo.gif
 Dynamic – Host your dynamic application
behind the CDN. For ex, Edgecast and
Panther Express.
 Very cheap - $0.2/GB

How Dynamic CDN works

www.yoursite.com

How Dynamic CDN works

www.yoursite.com

Static content
cached,
Compressed
automatically

13 disasters for production websites
1. Faulty hard drive supplied by supplier, data corruption within weeks.
2. Controller malfunctions and corrupts all disks in the same controller.
3. RAID malfunction.
4. CPU overheated and burned out.
5. Firewall went down.
6. Remote Desktop stopped working after a patch installation.
7. Remote Desktop max connection exceeded. Cannot login anymore to
servers.
8. Database got corrupted while we were moving the production
database from one server to another over the network.
9. One developer deleted the production database accidentally while
doing routine work.
10. Support crew at hosting service formatted our running production
server instead of a corrupted server that we asked to format.
11. Windows got corrupted and was not working until we reinstalled.
12. DNS goes down. Don’t get domain from GoDaddy.
13. Internet backbone goes down in different part of the world.

http://tinyurl.com/omar13

Conclusion

 ASP.NET out of the box, does not scale for millions of
hits.
 Must make the hacks at code, database and
configuration level to get it to scale.
 That‟s reality for any technology, not ASP.NET specific.

Scaling asp.net websites to millions of users

Related slideshows

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

More Related Content

What's hot

What's hot (20)

Similar to Scaling asp.net websites to millions of users

Similar to Scaling asp.net websites to millions of users (20)

Recently uploaded

Recently uploaded (20)

Scaling asp.net websites to millions of users