Tweet
In my line of work, (enterprise storage support), I am constantly amazed at how many hundreds of thousands of dollars worth of gear our customers will buy without having the least clue how they work.
Common scenario for me: One of my companies products does remote storage replication. You have one box in one site, the other somewhere else (sometimes thousands of miles away)... all I/O to Box A is replicated to Box B. This is a pretty standard feature, but it is now available in boxes much cheaper than they used to be.
All this wonderous niftiness requires cluster overhead, plus protocol overhead, plus the actual bandwidth you need to perform the data transfer. So, if you write to your database with peaks at 20MBps at site A, you will need around a 23-25MBps pipe, at least.
Storage guy goes to his networking guy and asks for the appropriate WAN bandwidth, which then gets attached to an overpriced converter box so it can talk to the storage. Much time passes and much money is spent, and the line is installed and everybody is happy... right?
Nope, it doesn't work. Much whining and escalations later it ends up in my queue. I look at the logs and note, just as everybody else has, that there are timeouts out the wazoo. I ask Storage guy what he is using to provide the long-haul link. He proudly replies: "A pair of T-1's."
*bangs head on desk*
A pair of T-1's is a measly 3Mbps (or 0.3MBps). He barely has enough bandwidth for cluster overhead, much less his actual workload.
Reply: "but... but... he said it was FAST! It already costs thousands of dollars per month! Why won't it work? MAKE IT WORK!"
*sigh*
SirWired
Common scenario for me: One of my companies products does remote storage replication. You have one box in one site, the other somewhere else (sometimes thousands of miles away)... all I/O to Box A is replicated to Box B. This is a pretty standard feature, but it is now available in boxes much cheaper than they used to be.
All this wonderous niftiness requires cluster overhead, plus protocol overhead, plus the actual bandwidth you need to perform the data transfer. So, if you write to your database with peaks at 20MBps at site A, you will need around a 23-25MBps pipe, at least.
Storage guy goes to his networking guy and asks for the appropriate WAN bandwidth, which then gets attached to an overpriced converter box so it can talk to the storage. Much time passes and much money is spent, and the line is installed and everybody is happy... right?
Nope, it doesn't work. Much whining and escalations later it ends up in my queue. I look at the logs and note, just as everybody else has, that there are timeouts out the wazoo. I ask Storage guy what he is using to provide the long-haul link. He proudly replies: "A pair of T-1's."
*bangs head on desk*
A pair of T-1's is a measly 3Mbps (or 0.3MBps). He barely has enough bandwidth for cluster overhead, much less his actual workload.
Reply: "but... but... he said it was FAST! It already costs thousands of dollars per month! Why won't it work? MAKE IT WORK!"
*sigh*
SirWired
I AM the evil bastard!
Comment