>> "mid-range setup" (iCore/xeon + DDR3)
The distinction here is app server class machines and database server classes machines.
app servers typically have less RAM, less I/O bandwidth, less disk optimization, and also may have to share hardware. Any virtualization matters - some virtualization slows the I/O, some, like para-virtualization, does not.
I've come across this a couple of times now - organisations not running a graph store on a database server class machine. Graph stores are databases.
On 11/12/17 10:31, Dick Murray wrote:
Consider the simplified steps involved in getting a single quad into a store (please correct me Andy);
That's right.
Read quad from source. Verify GSPO lexical and type.
For each slot, get the NodeId, creating it, and indexing it, if needed.
Check GSPO for uniqueness (read and compare) possibly x4 write to node->id lookup.
Doing the indexes at the same time is bad.
Write indexes.
Writing is cached.
Repeat n times. Understand binary format and tweak appropriately for tdbloader2 ;-)
