Messages in this thread | | | From | "Stephen C. Tweedie" <> | Date | Thu, 10 Feb 2000 14:34:58 +0000 (GMT) | Subject | Re: [RFC] 2.3.39 zone balancing |
| |
Hi,
On Thu, 10 Feb 2000 00:19:53 +0100 (CET), Andrea Arcangeli <andrea@suse.de> said:
> We just have it as far I can tell: zone->pages_{min,low,high} and it has > to be compared with the classfree value of each zone. Really I have not > checked the whole implementation line by line so I am not 100% sure > classzone does what I want.
> Actually I think freepages.* for kswapd can be completly replaced by > takeing the NORMAL _class_zone over the pages_high watermark of the > NORMAL zone.
Agreed --- the freepages thresholds are primarily there for atomic requests, and those usually come from the normal zone.
However, we still want to allow some hysteresis between page allocations and the page stealer so that we're not continually waking up kswapd just to get a single page back. The old behaviour where we disable kswapd once freemem goes up to the low-watermark and only reenable it once it gets to the mid-watermark handles that side of things well. It's probably less important to give the user control over those thresholds than it is for the GFP_ATOMIC thresholds, though.
--Stephen
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |