Bcache looks great if you don't lose power. If you do, I hope they did a *very* carefully thought out implementation.
See this article for why: http://www.cse.ohio-state.edu/~qin/pub-papers/SSDFault-FAST13.pdf On Thu, Mar 20, 2014 at 2:28 PM, Ishaaq Chandy <[email protected]> wrote: > Interesting. Especially since at work we've been leaning towards using > bcache for performance reasons to be able to deal with the flood of input > we get - not so much for ZooKeeper but for Cassandra. Do you have any > opinions about bcache? > > http://bcache.evilpiepirate.org/ > > > On 21 March 2014 07:14, Ted Dunning <[email protected]> wrote: > > > SSD's also have the issue that it is common that recently written data is > > not actually persisted. Worse, new data might be persisted while > slightly > > older data is not. These issues differ greatly across different > hardware. > > > > Disks with write caching disabled are vastly better understood. > > > > > > > > > > On Thu, Mar 20, 2014 at 1:09 PM, Patrick Hunt <[email protected]> wrote: > > > > > My experience with SSDs has been negative. Write cliff issues > > > eventually kick in and everything stops (if you put the txnlog on > > > there). See my earlier messages to the list about this. > > > > > > Patrick > > > > > > On Thu, Mar 20, 2014 at 10:14 AM, Software Dev > > > <[email protected]> wrote: > > > > I was thinking SSD for zookeeper but traditional for the log > directory. > > > > Memory wouldn't be a problem > > > > > > > > > > > > On Wed, Mar 19, 2014 at 11:14 PM, Michi Mutsuzaki < > > [email protected] > > > >wrote: > > > > > > > >> It should be fine to consolidate so long as these applications don't > > > >> overload the ZooKeeper cluster in terms of memory usage and > read/write > > > >> throughput. I would definitely test it first though :) > > > >> > > > >> On Wed, Mar 19, 2014 at 9:57 PM, Software Dev < > > > [email protected]> > > > >> wrote: > > > >> > We currently have 4 separate ZK clusters (hbase, kafka, solr > cloud, > > > >> storm) > > > >> > with either 3 or 5 per cluster. Should we combine all clusters > into > > > one > > > >> and > > > >> > just serve each one up in their own chroot? > > > >> > > > > > >
