As IT continues to aggressively centralize services -- and increasingly into sets of hot-hot data centers -- the importance of that inter-data center network traffic increases. Thanks to increases in backup traffic, virtual machine (VM) migration, service-oriented architecture (SOA) or replication among storage arrays, enterprises are generating greater volumes of traffic between data centers.
At the same time, enterprise expectations for performance are increasing too. Businesses expect quick backups, synchronous replication and responsive applications no matter what activity is happening between data centers.
When performance problems crop up and inter-data center communication is fingered as the culprit, the discussion inevitably turns to adding WAN optimization on those links. Adding inter-data center optimization to the network is not the only choice, however. IT staff should evaluate other options before deploying a WAN optimizer.
Here are a few tips for improving performance and WAN link utilization between data centers without buying an inter-data center optimization appliance or software:
Deduplicate storage. One particularly vexing class of performance problems stems from replicating storage among data centers and/or backing up data across WAN links. In either case, deduplicating storage can lead to significant improvements in performance by dramatically reducing the amount of data that moves. As an added benefit, data deduplication can also reduce storage consumption. Several data center managers Nemertes has interviewed have said that they were able to avoid or drop inter-data center optimization once they had robust data deduplication in place.
Update protocols and use all features. Many enterprises still use old versions of core applications and WAN protocols even though newer versions can provide greatly improved performance. Other times, they use protocols in the simplest way, failing to take advantage of features that can reduce inter-data center traffic burdens.
A best case in point is moving from the Server Message Block (SMB) protocol to SMBv2. Moving to the post-Vista versions of Microsoft's file sharing protocol takes a lot of round-trips out of the picture. This hugely improves performance across high-latency links while somewhat reducing the overhead -- or additional volume -- associated with getting a given amount of content replicated across such links.
Other protocols also offer options for reducing traffic volumes, such as those with built-in compression features. For example, SOA participants communicating via Simple Object Access Protocol (SOAP) -- sometimes generating multi-megabyte XML messages -- have the option of doing compression themselves; this can reduce data volumes by as much as 90%.
When applications are developed in-house, IT should push for development standards that make efficient use of network bandwidth through good application design. As an example, you can speed embedded SOA traffic this way: If you retrieve data from Service Component A in Data Center 1 and send it to Service Component B in Data Center 2, then you can embed that data into the traffic of Component A sent to B. However, if the database that the data is being pulled from is replicated between the data centers, you can have Component B pull data from the local copy; then, instead of the data going across twice, the data is only transported once.
Get QoS. Lastly, inter-data center performance improvement can be achieved through Quality of Service (QoS) features on existing infrastructure. Network staff can use the Differentiated Services Code Point (DSCP) portion of IP packet headers to tag packets belonging to specific classes of traffic by working with servers' network stacks, routers, switches, other network appliances and service providers (as needed). Different values of the DSCP can be used to indicate priority delivery. For example, DSCP can ensure that storage replication doesn't wait while email is being transferred among mail servers; it can also ensure packets aren't dropping where retransmits would make application performance unacceptable, which is especially important for real-time protocols like Voice over IP (VoIP).
The bottom line is deploying a WAN optimization appliance or software is not the only path to application delivery optimization. Paying attention to other layers of the infrastructure and other methods of improving performance can have just as profound an effect in optimizing WAN traffic between data centers.