Networking 101
Nov
17
Written by:
11/17/2009 4:34 PM
I think that I may have to smash my head into a steel door the next time someone tries to argue network performance with me. I have had this conversation WAY too often for 15 years and can’t take it anymore. I have to blog about it in the hopes of performing some enlightenment. There are way too many people that try arguing that their ping latency is next to nothing, and they usually debate that their WAN is perfectly stable. Trust me, you’re wrong, wrong, wrong. Please allow me to explain why.
PING MEANS CONNECTIVITY – I am SO tired of people telling that they have great WAN connectivity between their sites with excellent ping times. Ping s a very poor indicator of data transfer latency because it runs as an ICMP protocol on IP, and has high priority. Getting a ping response from another machine means that it’s on the network, nothing more can be reliably discerned. One of these days I’ll write a little utility that uses TCP and UDP to send user definable data payloads to another machine and measure that latency.
“OUR WAN NEVER GOES DOWN” – What you don’t know WILL hurt you and you won’t even know it. WAN’s are constantly dropping connection between sites. You just never really notice it because the vast majority of data transmission is bursty, so the loss of one packet means that you might get a red ‘X’ where a graphic should be. Hit refresh and you’re on your way again. Just you try doing that with database replication or off-site copy and you’ll cry your eyes out when you lose your production machine and find out that the replica database is suspect and won’t come on line because of the corruption. Woops… That’s specifically why Double-Take Software patented their data integrity algorithms back in the mid-90’s, to guarantee data integrity with transactional awareness. We have customers with latencies measure in tens of minutes and even months (think about that one for a while).