In the last episode (Jul 27), Mike Spreitzer said:
> Thanks for the clues. In this case the storage is not on a SATA disk,
> rather is it on a GPFS (
> http://en.wikipedia.org/wiki/IBM_General_Parallel_File_System) mount. This
> thing is capable of quite a lot more I/O bandwidth. I invoked `wc` on a
> large file and it took the "bi" stat of `vmstat` over 16000. The iostat
> utility is not installed there, I will look into that.
wc is sequential (and likely is reading very small chunks at a time, no more
than 8k, which further inflates your sequential iops number), while mysql is
doing random access on that table. Single-threaded random access is always
equal to the _average_ iops of all disks, not the total. It doesn't matter
what filesystem you use, since you can't predict random access and each read
is going to move one of your drive heads (but you don't know which one).
> Regarding your suggestions how to improve, I suspect I will not be able to
> put fldrcv in a RAM table. "SHOW TABLE STATUS" says the data_length of
> fldrcv is about 2 GB. I will look into your other suggestions.
You did say you had 64GB of memory? I wouldn't expect 2gb to be noticed at