Do you know what’s happening in your data center? I’ve been working with two customers who are just starting to monitor the switch ports within their data centers. There are many good reasons for complete data center monitoring of all interfaces/ports.
- Most data centers employ some amount of oversubscription, particularly if 10Gbps server links are used. Using vPC or VSS (or the upcoming TRILL) supports aggregation of multiple uplinks (relative to Spanning Tree), but there is still oversubscription in all but the most lavishly built data centers. Collecting and trending performance and error data on server ports and switch uplinks can provide advance notification that uplinks are becoming saturated. Servers that are generating the load when the uplink saturates can often be moved (at least in a VM environment) to other VM hosts where there is available uplink capacity. Or perhaps the workload can be time-shifted so as to even out the network traffic load.
- Most data center traffic these days is east-west (server-to-server), making it important for data center monitoring to identify problems before they impact the business. Flow data or packet captures are typically needed to identify these traffic flows. Uplink loads can be reduced by locating dependent servers to the same switches, which takes advantage of higher backplane speeds than are available through uplinks. Just make sure that the high-availability requirements are not compromised in the process. Being able to provision adequate bandwidth between servers running key applications means that those applications are running at peak efficiency, which means that the productivity of the staff using those applications is not impacted.
- Look for server ports with extremely low utilization over long periods of time. Depending on the business, the period may need to be as long as 90 days or more, to identify servers that are used once a quarter or less. Some of the low utilization servers may no longer be actively used. Identifying these servers allows the VM resources to be re-allocated to new services. Don’t forget that backups are often running on these servers, increasing the volume of traffic that must transit the network during backup periods.
The above items also help prepare for a transition to virtualization as well as on-going monitoring after the transition. A combination of interface utilization monitoring, along with flow data collection and analysis, can help you make sure that the data center is operating correctly and efficiently. Congested links can be easily identified — often before the applications are impacted.
What are you doing for data center monitoring?
-Terry
_____________________________________________________________________________________________
Re-posted with Permission
NetCraftsmen would like to acknowledge Infoblox for their permission to re-post this article which originally appeared in the Applied Infrastructure blog under http://www.infoblox.com/en/communities/blogs.html
