Sorry, I noticed that this didn't get cc'ed to the list.
On Tue, 1 Aug 2000, Chris wrote:
> > Uhm. I have to say I'd like to know a lot more about
> > your configuration,
> > etc. For most of us, indexing in 3.2 is ~2x slower and
> > searches are about
> > 2x faster, unless you're comparing phrase searches in
> > 3.2 to non-phrase
> > searches in 3.1.x.
>
> Well, now I'm confused. Prior to me opening my mouth, I had run
> 3 separate tests. Both single words and multiple words (no
> phrase searching) for any matching word. Each time, the searches
> took 20-25 seconds.
>
> After seeing this response, I headed back in to start timing
> various search methods. To my suprise, response was "instant",
> both using single search words and multiple (Or's) search words.
>
> This made me sit back and think a bit.... I run the dig every
> saturday night, and it takes roughly 6-8 hours. The number of
> searches ran through the remainder of the weekend is VERY light.
> Every time I've tested it in the past, it was right after the
> dig, etc. was completed, or early monday when I've come to work.
>
> Right now, watching system load, the searches happen so darned
> fast that I can't catch them with a TOP. In contrast, when
> testing soon after the dig, the search (when it takes 20 seconds)
> may suck 35-50 percent of my processing.
>
> Is it possible that what I'm seeing is that much of the indexes
> have to be read from the disk for a while until much of their
> contents end up in ram???? That's the only thing that appears to
> explain the huge difference between right now and doing the
> searches immediately after a dig....
-(previous message)-----
So if I can summarize what you're seeing:
1. Indexing speed is fine. (No noticeable difference from 3.1.x)
2. Search speed soon after re-inedxing is slow.
3. After a day or so, search speed is lightning fast. (Faster than 3.1.x)
Do I have that about right?
Point 1 is a nice report, esp. in contrast to the flames I sometimes get
about indexing speed being slow. Point 3 is about what I expect for search
speed in 3.2. Point 2 is the curious one...
Yes, you're right that most systems will cache large parts of the
databases. Even with 3.1.5, I find that a follow-up search is much faster
than the original since parts of the DB are cached by the OS. But how big
are your DB? Why does it take so long for caching to hit?
-Geoff
------------------------------------
To unsubscribe from the htdig3-dev mailing list, send a message to
[EMAIL PROTECTED]
You will receive a message to confirm this.
