Transparent Caching The art of caching network traffic without requiring user / browser side configuration.

Transparent Caching

The art of caching network traffic without requiring user / browser side

configuration.

Who am I?

Overview

• What is transparent caching, and why use it?

• Tools available

• How to set it up

• Common problems

• Alternatives

What is transparent caching?

• Transparently proxying / caching network traffic without requiring user configuration or knowledge.

• A way to simplify caching for the end user

• Forces all users to use the cache.

Why use transparent caching?

• Ease of use. No configuration required by the end user.

• Catching all users. No users can bypass the cache.

Reasons to not use it

• It is not a magical tool without problems.

• Technical issues– Networking issues. “Woodo magic”– Stability / Reliability– Only port 80– FTP not supported– To be efficient modern browsers are required

Reasons to not use it (cont.)

• Political reasons– What is an internet connection?– Privacy

• No user control. Users can’t bypass the cache.

What is involved?

• TCP level routing

• Reverse NAT or related technology to hijack port 80 traffic.

• A proxy with some knowledge of transparent proxying

• A cache

Tools available

• TCP level Routing– Policy routing / route maps– TCP / layer 4 switches

• with or without NAT

– Cisco WCCP

• Host level NAT– Linux firewall code– FreeBSD firewall code– IP-Filter

Using policy routing to redirect traffic

• A standard router configured to route TCP port 80 to the cache server.

Router

Cache Server

Internet

User 1

User 2

User ..

User 3

Policy routing (cont.)

• Benefits– Can usually be deployed without extra hardware

• Drawbacks– Only static routing– No fault tolerance. Port 80 traffic disrupted if cache

server fails.– More CPU load on the router

Running the cache on a router

• Small network / firewall. Host used as router.

Router / Cache Server

Internet

User 1

User 2

User ..

User 3

Caching router / firewall

• Benefits– Less hardware required– Well suited for small to medium sized firewalls.

• Drawbacks– Stability / reliability. Can disrupt all

communication.– If running on a firewall: make sure the firewall

protects the cache software.

Cisco Web Cache ControlProtocol (WCCP)

• Developed by Cisco for Cisco

• Automated configuration. Proxy servers announce their presence to the router.

• Load balancing

• Fault recovery

• Commercial Licensing required. Not currently an option for free software.

TCP level / layer 4 switching

• The use of a smart and efficient network device to redirect traffic.

CacheServer

Internet

Router TCP SwitchUser 1

User 2

User ..

User 3

...

TCP switch benefits

– Can bypass the cache if it malfunctions– Good reliability– Can distribute the load on multiple cache servers– Can do the required NAT, allowing the use of any

OS on the cache server.– Some do HTTP proxy translation, allowing the use

of any proxy software.

TCP switch drawbacks

– One more expensive box to purchase– Using NAT requires switch vendor support in the

proxy software to support old browsers.

Request formats

• Proxy requestTCP connection from client to proxy

GET http://www.example.com/file HTTP/1.0...

• Server requestTCP connection from client to server IP

GET /path/to/file HTTP/1.0Host: www.example.com (if supported)...

Problems related to request formats

• A transparent proxy must reconstruct the URL of the request.

• Host: headers not always available. HTTP/1.1 feature or 1.0 add-on.

• IP address from NAT translation.

What happens at the TCP level?

• Normal communication / proxying– IP based routing– TCP is end-to-end– One IP address, one Host

• Transparent proxying– TCP based routing– TCP is no longer end-to-end– One IP address, “multiple hosts”

Problems at the TCP level

• TCP normally relies on two IP protocols. TCP and ICMP. Of these only TCP can be reliably redirected.

• ICMP is required for Path MTU discovery.

• TCP resets if a single packet travels another path bypassing the redirection.

Things to consider when configuring OS level NAT

• Try not to disturb traffic to/from the cache server host.

• Make sure that the proxy traffic is not redirected back to the proxy.

• Be prepared to do packet level traces, preferably from a separate box.

Recommended steps when building a transparent proxy

• Think it over. Is it really required?

• Build and test the proxy server

• Configure NAT on the proxy server

• Test it using a local LAN client

• Set up TCP level routing.

Common problems

• Communication hangs for some users– Most likely caused by MTU related problems.

• Connection reset errors– Usually misconfigured NAT or TCP routing.

• Bad performance– Possibly CPU bottleneck in the router.

TCP Reset errors (cont.)

• Error messages seen from the proxy.– TCP routing or NAT affects traffic generated by the

proxy.

• Error messages seen by the browser (popup)– TCP routing or NAT failure, causing some client

traffic to bypass the redirection.

Alternatives

• PAC files

• Blocking port 80– Selectively or everything– Possibly with an automated message

Selectively blocking port 80 with a message

• A good alternative to transparent proxying

• Uses the same techniques as transparent proxying for hijacking port 80, but only to deliver the instructions.

Blocking port 80, benefits

• Forces the users to configure their proxy settings

• Users are automatically provided with configuration instructions when needed. Less calls to support line.

• Users get the information on why caching is good for them.

• PAC file allows easy configuration of exceptions

Blocking port 80, drawbacks

• Not all browsers supports proxy settings

• Users are required to be capable of following instructions.

Summary

• Transparent caching is a good tool in most configurations to ease user side configuration.

• It has some important limitations. Not a full replacement for standard proxying.

• For many automatic instructions on how to configure proxy settings achieves the same goals.

Sources for more information

– Squid FAQhttp://squid.nlanr.net/Squid/FAQ/

– Router manuals on policy routing– IP-Filter home page

http://cheops.anu.edu.au/~avalon/ip-filter.html

– Linux 2.0 ipfadmhttp://www.xos.nl/linux/ipfwadm/

– Linux 2.2 ipchainshttp://www.rustcorp.com/linux/ipchains/

Questions

Example Cisco IP policy route map

• Policy route map, routing port 80 (www) to server 10.11.12.13

! Enable policy routinginterface Ethernet0 ip policy route-map proxy-redirect! Route to proxy serverroute-map proxy-redirect permit 10 match ip address 110 set ip next-hop 10.11.12.13! Only policy route client www trafficaccess-list 110 deny tcp any any neq wwwaccess-list 110 deny tcp host 10.11.12.13 anyaccess-list 110 permit tcp any any

Example Linux ipfwadm NAT

• Linux 2.0 redirecting eth0 TCP port 80 to Squid on port 3128– Kernel options:

CONFIG_IP_FIREWALL=y

CONFIG_IP_ALWAYS_DEFRAG=y

– ipfwadm ruleset# Accept local traffic

ipfwadm -I -a accept -W eth0 -D this.host

# Redirect port 80 to Squid on 3128

ipfwadm -I -a accept -W eth0 -P tcp -D 0.0.0.0/0 80 -r 3128

Example Linux ipchains NAT

• Linux 2.2 redirecting eth0 TCP port 80 to Squid on port 3128– Kernel options:

CONFIG_IP_FIREWALL=y

CONFIG_IP_ALWAYS_DEFRAG=y

– ipchains ruleset# Accept local traffic

ipchains -A input -j ACCEPT -i eth0 -d 10.11.12.13/32

# Redirect port 80 to Squid on port 3128

ipchains -A input -j REDIRECT 3128 -i eth0 -p tcp -d 0.0.0.0/0 80

Example IP-Filter NAT

• ipnat ruleset redirecting TCP port 80 to Squid on port 3128

# Redirect direct web traffic to local web server.rdr de0 10.11.13.13/32 port 80 -> 127.0.0.1 port 80 tcp# Redirect everything else to squid on port 3128rdr de0 0.0.0.0/0 port 80 -> 127.0.0.1 port 3128 tcp

Running Squid on Linux

What is Linux

• Linux is like any other UNIX

• POSIX standards

• GNU tools

• Best of SysV and BSD families

Filesystem performance

• To few performance counters for I/O to make any good measurements

• Asynchronous writes by default (like fastfs on Solaris)

• noatime mount option

Kernel performance / tuning

• Memory freelist tuning on smaller systems– /proc/sys/vm/freepages

• Filedescriptor limits– Default 256

– Later revisions of 2.2 may allow 1024– Patches available for higher limits

Hands on transparent caching

• Linux configuration– Kernel configuration

• Firewalling & Transparent proxy support

– ipfwadm configurationipfwadm -I accept -D thishost

ipfwadm -I accept -P tcp -D 0.0.0.0/0 80 -r 3128

Hands on transparent caching (cont.)

• Squid configurationhttpd_accel_host virtual

httpd_accel_uses_host_header on