On Mon, Jun 08, 2009 at 09:20:23AM -0400, Rik van Riel wrote: > Mel Gorman wrote: > >> We've encountered this before and the conclusion was that the current >> adjustments for watermark calculations of high-order allocations is right, >> or at least there is no better alternative. In other words, the page >> allocator in this instance is behaving as expected. Do we want to >> revisit that discussion as to whether the watermark calculations for >> high-order allocation should change? I think we'll reach the same >> conclusion or at least decide that allowing the order-1 atomic >> allocation to succeed here would just postpone the problem. > > It would not just postpone the problem, it would also > bring the system closer to a state where kswapd does > something about the order-1 free areas. > > This might postpone the problem indefinately. > How do you figure it does not just postpone the problem? If there are a batch of order-1 allocations that come in like this, it will eventually deplete the higher-order pages and then fail because kswapd is not getting woken up. Minimally, if we were to ignore the watermarks, there would need to be logic that says "If a high-order allocation would fail due to high-order watermarks not being met, but the watermarks are ok from an order-0 perspective and the high-order page is available, then grant the allocation but wake up kswapd as if the order-1 allocation had failed to get the high-order watermarks back in shape" > Currently the system fails early, without kswapd > kicking in and freeing new order-1 areas. > If the allocation was granted, then kswapd will still not kick in. -- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab -- To unsubscribe from this list: send the line "unsubscribe kernel-testers" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
