[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Feature request: Time to completion and checkpointing
From: |
Bruno Haible |
Subject: |
Re: Feature request: Time to completion and checkpointing |
Date: |
Tue, 20 Nov 2007 16:43:54 +0100 |
User-agent: |
KMail/1.5.4 |
Cem Karan wrote:
> I'd like to make a feature request; if the algorithm is amenable to it, it'd
> be nice if gperf could output an estimated time to completion for any given
> run.
This is hardly implementable: The algorithm perform many rounds of searches;
some rounds are very quick because the first tried value already fits; some
others are extremely CPU intensive.
You can get a very vague estimation of the time needed by running gperf on
a sample 1/10th or 1/2 of the size of your real needs. For example, when
I run gperf on the first 100 resp. 1000 lines of your sample, I get running
times around 0.1 sec resp. 15 sec. But yours has more than 300000 lines!
Now do the extrapolation...
> Even better would be some kind of automatic checkpointing, so that if
> a long run gets interrupted, gperf can restart at the last checkpoint.
Most users will consider gperf useless for a certain problem if it takes
more than 1 hour. In this case, normal developers will revert to a binary
search algorithm across a sorted array.
Bruno