We are going to aggregate Network Interface Card (NIC) 1 GB ports in order to get a faster connection between servers. A better aggregate throughput is just one of the advantages. Another advantage is redundancy. If one interface goes down, the other interfaces in the aggregate will still work. This is very similar to bonding NICs. We will be using HP Auto Port Aggregation (APA) software in combination with the Cisco PAgP protocol.
From the HP document APA Support Guide at http://www.docs.hp.com :
"HP Auto Port Aggregation (APA) is a software product that creates link aggregates, often called 'trunks,' which provide a logical grouping of two or more physical ports into a single 'fat pipe.' This port arrangement provides more data bandwidth than would otherwise be available. Two additional features are automatic link failure detection and recovery; it also offers optional support for load balancing of network traffic across all of the links in the aggregation. This enables you to build large bandwidth "logical" links into the server that are highly available and completely transparent to the client and server applications."
Now, how does this help our Oracle configurations? Simple, I will use the aggregated ports as the host for a database link between servers. This way, we will be able to move data faster between Oracle databases.
Our plan is to aggregate 4 ports which will give us the capability to transfer between 3.7 and 3.9 GBbit/sec. APA supports up to 8 aggregated ports. When you go over 2 aggregated ports, you begin to see diminishing returns which is why we will not get a total throughput of 4 GBit/sec.
1 comment:
Here is something you need to know. All of the HP material makes it sound like that creating an APA interface will increase your overall bandwidth. Don't drink the cool-aid. It doesn't. What you really end up with is a single ip address that maps to x number of cards. You still have to submit several jobs to utilize all of the cards. So, the only thing that you gain from APA is better reliability and failover because a single nic failure can be tolerated. That is the bottom line. It doesn't stripe your transfers across all of the links it uses a round-robin approach.
Post a Comment