Hmm. That "device about to die" write latency signature is interesting. I have pinged our hosting company asking for specifics as to exactly what USB thumb drive they supply.
On Sat, Nov 16, 2013 at 6:30 PM, Dan Simpson <dan.simp...@gmail.com> wrote: > It doesn't seem like a great idea. The USB drives typically use dynamic > wear leveling. See this analysis on wear: > https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&ved=0CD8QFjAC&url=http%3A%2F%2Fwww.usenix.org%2Fevent%2Ffast10%2Ftech%2Ffull_papers%2Fboboila.pdf&ei=qZyHUrizFtKAygGs9YGoCg&usg=AFQjCNHTC7d6fcI1CNWmjbHMwgXI1nUWcQ&sig2=BaWgHj3ib-cQOBPQsoCadA&bvm=bv.56643336,d.aWc&cad=rjt > > If you you do end up using it, make sure to monitor write latency so you > don't get hit by the bus. > > > On Sat, Nov 16, 2013 at 6:12 AM, David Tinker <david.tin...@gmail.com> > wrote: >> >> Our hosting provider has a cost effective server with 2 x 4TB disks >> with a 16G (or 64G) USB thumb drive option. Would it make sense to put >> the Cassandra commit log on the USB thumb disk and use RAID0 to use >> both 4TB disks for data (and Ubuntu 12.04)? >> >> Anyone know how long USB flash disks last when use for a write heavy >> workload like this? >> >> Please tell me if this is a really bad idea. >> >> Our alternative is to use one 4TB disk for commit log and one for >> data. Of course this will give us only half the space. >> >> Thanks >> David > > -- http://qdb.io/ Persistent Message Queues With Replay and #RabbitMQ Integration
