Re: orion

Kevin Closson wrote:

> >>>
>
>
>>>>1. run simultaneously from each node. Will this model RAC?
>>>>or does orion always read/write from the same part of the
>>>>disk. Does orion play nice when overlapped like this? Will
>>>>it be better if the tests are staggered?
>>>>
>>>>
>
>
>The offsets are not coordinated. That doesn't really matter
>that much though, really. Just make your fileset so large
>that it reduces the likelihood of a clobber. But, honestly,
>having writes to the same file offset from 2 nodes really
>doesn't invalidate this workload.
>
>
>--
>http://www.freelists.org/webpage/oracle-l
>
>
>
>
>

Thanks. I tried that and it seemed to work as expected (plotting iops(node1)+iops(node2) vs load has similar characteristics to iops(node1) run by itself. I just need to double the load value on the x-axis when summing the two nodes). This is with a quick plot, I need to do it a bit more carefully to be sure. The two observations that jumped out on a first scan of my data is that: write latency is about an order of magnitude faster than read latency, iops for 100% read is pretty much flat when plotted vs load. I am wondering if the latency result is a manifistation of orion. Does it not care where the writes go? just put them on disk and don't wait for any head movement or rotation while the read has that latency? or is there more?

If I find anything interesting here I'll post.

Henry

--
http://www.freelists.org/webpage/oracle-l