Database Reference
In-Depth Information
Load balancers
Most of the enterprise implementations of web services will demand high availability.
In some cases it will be a global high availability that is usually achieved via the sys-
tem architecture's designs and deployments leveraging multi sites in active-active or
active-passive. The passive site is either standby (manual process to fail over and fail
back) or hot standby (automatic process to fail over and fail back) to provide the ser-
vices if the primary site goes down. The load balancers are also used to achieve the
horizontal scalability. Load balancers for web services are implemented at either L7
(application Layer) also known as
Global Server Load Balancer
(
GSLB
) or
Global
Traffic Manager
(
GTM
), or at L4 (Transport Layer) also known as
Local Server Load
Balancer
(
LSLB
) or
Local Traffic Manager
(
LTM
) to achieve high availability and
scalability. Layer 7 forwards the user to the appropriate local load balancer, any fur-
ther requests for the user will directly go the local load balancer within a specified time
window configured as part of
TTL
(
Time to Live
). The Layer 4 load balancer serves
all user requests. In other words, the L7 load balancers provides the user forwarding
service while L4 load balancers serves the users as middlemen. In a high-availability
environment, both L4 and L7 are used. These load balancers come in both software
and/or hardware appliances from various vendors. The major vendors are F5, Broca-
de, NetScaler, and Cisco.
L4 load balancers are usually implemented to achieve high availability within a data
center that provides limited horizontal scalability. One needs to implement L7 load
balancers to achieve high horizontal scalability. You will mostly find that L4 and L7
load balancers are implemented as a pair, since most implementations don't need to
scale beyond a single load balancer capacity. In implementations where L7 load bal-
ancers are not used; the L4 load balancers pair is implemented with one of them as
hot standby. It is recommended that if you have got L7 load balancers then make all
the L4 load balancers active to optimize the performance and capacity during peak
loads.
L7 load balancers/GSLB/ GTM configurations are as follows:
• Implement L7 health checks over TCP health checks
• Use low TTL such as 30 secs
• Utilize GEO affinity, if available
• Implement independent endpoints if services are run by different server
processes
