Coyote Point Systems Equalizer GX Series, Administration Manual

Page: 859 / 866

Persistence
Often, when a client (web browser) connects to an application, there is some "shared state" between the client and server which
cannot be used with any other server. Using the following example: You point your web browser at www.website.com and log in.
The server notes that you are logged in and allows you to use the application. Now,for example, there were two servers and a load
balancer, and that the load balancer just alternated handing any requests it recieved between the two servers... You've been con-
nected to server "A" where you log in. You then send a different request (run a report) and the load balancer sends your request to
server "B". Since you are not logged into server "B" so you get an error and are asked to log in again. This is because you have NO
PERSISTENCE. Your connection does not get connected to the same server each time. What you need is some way for the load
balancer to recognize your browser and always send your requests to server "A". We do this in two ways. #1 is called "IP per-
sistence" or "sticky persistence". The load balancer remembers your IP address the first time you connect to a server and records
which server you were connected to. The next time you connect, Equalizer looks up your IP address in an internal table and notes
that you belong on server "A", so that is where you are sent. Sticky Persistence is available with both L4 and L7 load balancing. An
alternate persistence scheme is available under L7 only. It is called "cookie persistence" and it uses information stored at the client
(a cookie http://en.wikipedia.org/wiki/HTTP_cookie) to recall which server it should be connected to. The advantage of keeping the
persistence information on the client is that no memory is used on the Eq. This way millions of clients can have poersistence for an
indefinate period of time. With sticky persistence the number of persistent clients is limited by the memory available to store the
sticky table.
physical server
A machine located on the internal network that provides services on specific IP addresses and ports. See server and virtual web
server. See also authoritative name server, back-end server, name server, and proxy server.
piece
An atom followed by a single *, +, or ?, or by a bound. See atom, branch, and regular expression.
ping
A program used to test reachability of destinations by sending them an ICMP echo request and waiting for a reply. See echo and
probe. See also CMP echo request
port
The abstraction used by Internet transport protocols to distinguish among multiple simultaneous connections to a single destination
host.
port grouping
Refers to the configuration of a load balancers with a list of application ports that must be treated as one group.
port number
The number used to identify a service contact port, such as HTTP port 80.
Port-Address Translation (PAT)
PAT is inherent in load balancers and refers to tranlsating the port number in TCP/UDP packets.
primary Equalizer
The primary unit that handles requests. If the primary Equalizer fails, the backup unit replaces it. See also backup Equalizer and hot
backup.
probe
An action that obtains status information about a computer, network, or device. See geographic probe and ping.
probing interval
The target interval between TCP probes of a cluster that has been marked failing in the load balancing daemon’s internal tables. If
the server does not respond to strikeout threshold (see below) additional TCP probes after it is marked failing, then the server is
marked down. These additional probes are at least probe interval seconds apart. This value is solely a target; the monitoring pro-
cess adjusts itself based on a number of factors, including system load. The default value is 20 seconds.
Copyright © 2014 Coyote Point Systems, A Subsidiary of Fortinet, Inc.
All Rights Reserved. 859
Equalizer Administration Guide