The link to the mail discussion is broken but here it is: http://wiki.apache.org/hadoop/MachineScaling
"Hadoop benefits greatly from ECC memory, which is not low-end, however using ECC memory is RECOMMENDED." Generally speaking you can redo a map reduce task if it faults due to ram issues, but you can't always redo a hbase operation. -ryan On Mon, Nov 29, 2010 at 11:58 PM, Christian van der Leeden <[email protected]> wrote: > Hi, > > I guess this is only reliability, since the read failures will be > corrected > but also reported, so you have a possibility to change the memory that is bad, > Without ECC you get garbage or read failures (SIG 11 e.g.) and your node > will be corrupted or down. > > Christian > > On Nov 30, 2010, at 5:31 AM, Hari Sreekumar wrote: > >> Hi, >> >> I read recently that ECC memory is better for HBase/Hadoop. But isn't ECC >> memory slightly slower than normal memory? How much extra reliability does >> ECC provide? Has anyone done a comparison of ECC vs normal memory stress >> test of HBase or hadoop in general? It also costs more. Is it worth it? >> >> Thanks, >> Hari > >
