For IP networks, little variation on the network is highly desirable to ensure predictable application performance as the workload varies.
Other reasons include:
• Low, deterministic latency
• Guaranteed delivery
• Smaller administrative domain, which is easier to troubleshoot
• Fewer configuration compromises
• Better fault isolation
• Less complexity to upgrade and maintain.
These reasons can be easily backed up when problems arise. When diagnosing technical problems, the smaller the domain of analysis, the easier it is to perform the analysis.
This supports the idea of a dedicated network for storage traffic.
Upgrading and maintaining a large network can also become complicated. It is easier to maintain smaller scale networks that are dedicated to storage traffic than to maintain a large complex network that handles comingled storage and general-purpose network traffic.
For larger environments, there are compelling advantages to having a separate physical network for IP storage traffic, regardless of whether you use block storage (iSCSI) or NAS storage (CIFS and NFS).
Congestion on the network can result in a ‘slow start’, as the application tries to determine the network performance which can take seconds before full flow is re-established after a congestion event which manifests itself as slow storage access or application performance degradation.
Low and deterministic latency is dramatically affected by small-duration traffic spikes resulting in dramatic decreases in the transaction rate for applications as they wait for storage I/O to complete.
TCP is required to achieve guaranteed delivery, but the side effects of the slow start mechanism can cause orders of magnitude increases in I/O latency for minutes at a time, dramatically slowing application response.
The process of diagnosing and eliminating this behaviour can be very complicated and time consuming.
Therefore, simplifying the data center network design by using a dedicated IP storage network for both block and NAS storage pays dividends.
Today’s IP storage applications are becoming increasingly complex, business-critical, and high-profile.
Failure is not an option, but running these bandwidth-intensive, latency sensitive applications on a shared network puts storage performance and availability at risk.
The best-practice solution for storage (FC or IP NAS) is a dedicated, fabric-based network, both inside and between data centres.
This is essential for maintaining reliable and predictable application performance, as well as providing security for data flows for storage between data centres.
Storage applications thrive on dedicated fabrics that are reliable, high-performing, self-healing, and easy to manage.
By Phil Coates – Systems Engineer Manager A/NZ, Brocade