iSCSI SAN Network Best Practices
I spoke to a customer earlier this week who had a small problem with an online capacity upgrade. He told me that while their new system and capacity were being integrated into their existing storage resources, performance dropped approximately 20%. I brought this up at the Atlanta Customer Advisory Board meeting today to see if the customers that were there had seen anything similar or had any comments. As it turned out, our local SE Wade O'Harrow had been there and gave us an overview. He said the customer's network was designed where the EqualLogic storage systems are connected to multiple edge switches in an core-edge network with relatively thin ISL uplinks between the edge and core switches. These links were not able to keep up with the combination of the customer's normal high-throughput I/O load and the additional load incurred when a new system is added to an existing group and the load is balanced among systems.
The comments from our customers on the advisory board were interesting. They said that core-edge topologies are part of commonly accepted best practices for Ethernet/TCP/IP networks, but that the assumptions used in forming those best practices did not necessarily apply to SANs - especially heavily loaded iSCSI SANs. Storage systems involve much higher traffic levels and I/O loads than most other end nodes connected to edge switches. Instead they offered that best practices for iSCSI SANs would connect storage at the core of the network as a centralized resource or that a different, beefed up edge design including fatter pipes were needed. They thought people would get a lot of value from better documentation of best practices for iSCSI SANs.
I agree. So, I'll try to revisit this topic regularly in this blog and let you know what I find out what's working for other customers.







