eBay Makes Huge Gains In Parallel Efficiency 47
CurtMonash writes "Parallel Efficiency is a simple metric that divides the actual work your parallel CPUs do by the sum of their total capacity. If you can get your parallel efficiency up, it's like getting free servers, free floor space, and some free power as well. eBay reports that it amazed even itself by increasing overall PE from 50% to 80% in about 6 months — across tens of thousands of servers. The secret sauce was data warehouse-based analytics. I.e., eBay instrumented its own network to do minute-by-minute status checks, then crunched the resulting data to find bottlenecks that needed removing. Obviously, savings are in the many millions of dollars. eBay has been offering some glimpses into its analytic efforts this year, and the PE savings are one of the most concrete examples they're offering to validate all this analytic cleverness."
Or maybe this was done to a mgmt goal? (Score:4, Insightful)
I can imagine this: Some eBay hotshot comes up and says: "we only use 50% of our servers, we've got to do better here". So:
1) They don't buy new servers. Workload increases, better utilization, no analytics involved.
2) Or, someone got clever, and added an idling process to each idle server. Presto, we've improved our PE -- and we've got a nice yearly bonus as well.
The article actually says nothing, besides claiming a supposed 1.6x improvement, besides a very vauge refrence to analytics. This ./ post is actually meant to promote www.xlmpp.com.
self promotion (Score:2, Insightful)
Slashvertisement (Score:5, Insightful)
Yes, we know eBay is trying to boost it's stock value and xlmpp.com wants more traffic.
Tag it what it is an move on. There's not much real info here.
But where's the beef? (Score:5, Insightful)
What interesting things did they learn? What were they doing wrong before and what did they change?
I don't see any hard facts or much useful info.
Car analogy: it's like Ford says we've improved engine efficiencies 1.6X.
But you don't even get new MPG figures, no comparison of 0-60 before and after (to show whether there was any impact on performance), no torque curves, not even a mention of "high intensity electric fields reducing viscosity".
So to me it's as good as some PR firm bullshit and should not be on Slashdot.
Heck it's about as much useful news as programmer productivity improving because Slashdot went down for a day.