Re: Load Distributed Checkpoints, take 3

On Sun, 24 Jun 2007, Simon Riggs wrote:

> I can't see why anyone would want to turn off smoothing: If they are
> doing many writes, then they will be effected by the sharp dive at
> checkpoint, which happens *every* checkpoint.

There are service-level agreement situations where a short and sharp
disruption is more acceptable than a more prolonged one. As some of the
overloaded I/O tests are starting to show, the LDC may be a backward step
for someone in that sort of environment.

I am not a fan of introducing a replacement feature based on what I
consider too limited testing, and I don't feel this one has been beat on
long yet enough to start pruning features that would allow better backward
compatibility/transitioning. I think that's introducing an unnecessary
risk to the design.

> We won't need to set checkpoint_segments so high, since performance is
> smoothed across checkpoints by LDC and its OK to allow them more
> frequently. So this concern need not apply with LDC.

Performance *should* be smoothed across by checkpoints by LDC and my
concern *may* not apply. I think assuming it will always help based on
the limited number of test results presented so far is extrapolation.

--
* Greg Smith gsmith(at)gregsmith(dot)com http://www.gregsmith.com Baltimore, MD