Application Delivery Controller class Web Compression, such as NetScaler’s, is asymmetric and decompressed at the far end by the web browser, which signals it can do in the HTTP Header AcceptEncoding. Non-browser rendered applications need a compression agent at the client for symmetric compression, provided by the WAN Optimization Controller class of product, like Citrix’s CloudBridge. WAN Optimization Controllers provide symmetric compression, but much more powerful is DeDuplication, or Data Redundancy Elimination. CloudBridge’s TCP Optimizations provide relief from latency and platforms sometimes reference the number of TCP Connections supported. Several applications get specific optimizations and can be securely unencrypted for treatment. Video caching in the branch, QoS, traffic management, and other cool features are also available in CloudBridge. The CloudBridge Technical Overview Whitepaper is a great reference. CloudBridges and NetScalers complement each other’s functionality and are often seen together for Citrix and many, many non-Citrix Applications like MAPI and encrypted MAPI for Exchange Email, HTTP, SnapMirror, and more… Some of the CloudBridge units benefit from the NetScaler Feature “Cloud Bridge Connector” and provide high scale application and TCP connection support with SDX and NetScaler.
Knowing the symmetric nature of the CloudBridge one will expect to have an instance in the branch, remote office, other Data Center, Main DC, mobile user PCs and Citrix Receiver on PC users. Really, each pair of locations with traffic flows between. When sizing a network for CloudBridge acceleration, we are looking for congested links and large latency numbers like 300ms to India, or 500ms to the International Space Station. Maybe it is 30ms to Location B, and 12 branch offices on E1, 2MB circuits from 60 to 120ms and ~70% utilization. We need to size a 2MB platform for 12 locations, and a 12x2MB or largest WAN Circuit – whichever is larger. So if all branches were going full on, that would be 24MB, but maybe they have a DS3 at 45MB. We need to consider sizing between these numbers. If you exceed your sized limit, the traffic passes through, but without optimization, compression, acceleration, or treatment. Do you want High Availability? Often the Appliance will close Network Interface Cards to bypass a failed unit, for again – unoptimized, pass through traffic, until the connections rebuild post failure. Other HA and interception methods should be considered. For sizing, we are mostly counting boxes. Consider also your DC to DC traffic, and can you overlap work hours? We need to count bandwidth for DC to DC as well.
How about AWS or Azure? CloudBridge Connector on NetScaler VPX can connect for a IAAS DC option. Need to count those flows in the sizing too.
In the CloudBridge DataSheet, one can select the correct size platform. If we look at the Bandwidth as our first sizing metric, one can figure the 12 sites of 2MB each would make a good CB600-002 or CB700-002. The TCP Connections per platform, Maximum HDX Sessions per Platform, and PC based client connections are also key metrics to consider. It is a recent general baseline estimate of 15 TCP connections per active user, and with replication and other DC to DC operations Port Usage is known. Customers could run real netstat research and figure there TCP connections, or perhaps use NetFlow or other reports to determine exact numbers if desired. One will want to consider the bandwidth first, then TCP connections, ICA Plugin or PC Client use, and HDX ICA sessions too.
At the Data Centers, one needs to consider the branches (remember 12×2) and the DC to DC, and other location traffic concurrent numbers. If I wanted to plan all branches might go at 100% circuit speed, then that is 24MB. I have a 45MB and full mesh MPLS, so the rest of the 45MB DS3 could be used at the same time for my other DC. Both DCs are 45MB attached in my example. They could do an optimized 21MB between them and the Main DC pipe would be full. 45MB DS3 with all active branches on it running at 100%.
What if you needed more? A CB2000-050 or CB3000-050might cover 45MB, and you could go up in steps, to the CB5000-2000 for 2GB if your accelerated traffic throughput is higher scale. One could go larger with WCCP Groups with ACL and interception filters with multiple instances.
Again, one needs to understand the use cases and applications. I always recommend running it all by the Citrix SE, and wanted to blog my thoughts on how it is done. Best wishes and happy sizing….