On 05/08/2014 10:50 AM, Kian-Tat Lim wrote:
> Is this the kind of hardware we'd want to specify for Qserv deployment?
>
> <http://www.supermicro.com/products/system/1U/6017/SYS-6017R-73THDP_.cfm>
>
Maybe. They're dual-socket for a CPU family that's 4-10 cores/socket, so 
there should be a good amount of CPU. Cache is up to 30MB, but if we are 
sustaining 200MB/s for each of 12 drives, then we are refilling cache at 
80/sec on data alone, so every 10ms or so. It could work.

I should probably have a look at the shared scan implementation--if we 
have 12 spindles (even if we bundle them in pairs-->6), CPU isn't 
infinite to the scheduler any more.

I'd really, really, rather have smaller nodes: single sockets, less 
drives per node. Our unit of failure is a node, so I'd rather do 
recovery in smaller, rather than larger pieces. "We just lost a node; 
there goes 100TB; wait, what?"

-Daniel

########################################################################
Use REPLY-ALL to reply to list

To unsubscribe from the QSERV-L list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1