From owner-freebsd-fs@freebsd.org Fri Jul 31 09:46:20 2015 Return-Path: Delivered-To: freebsd-fs@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id E5A159AF05C for ; Fri, 31 Jul 2015 09:46:20 +0000 (UTC) (envelope-from lev@FreeBSD.org) Received: from onlyone.friendlyhosting.spb.ru (onlyone.friendlyhosting.spb.ru [46.4.40.135]) by mx1.freebsd.org (Postfix) with ESMTP id A8F9D192A for ; Fri, 31 Jul 2015 09:46:20 +0000 (UTC) (envelope-from lev@FreeBSD.org) Received: from lion.home.serebryakov.spb.ru (unknown [IPv6:2001:470:923f:1:2924:7e01:7d9c:bbfe]) (Authenticated sender: lev@serebryakov.spb.ru) by onlyone.friendlyhosting.spb.ru (Postfix) with ESMTPSA id 42D6E2EDC for ; Fri, 31 Jul 2015 12:46:18 +0300 (MSK) Date: Fri, 31 Jul 2015 12:46:11 +0300 From: Lev Serebryakov Reply-To: lev@FreeBSD.org Organization: FreeBSD X-Priority: 3 (Normal) Message-ID: <575615785.20150731124611@serebryakov.spb.ru> To: freebsd-fs@freebsd.org Subject: ZFS with large blocks on 10.2-PRERELEASE r286065: still have big problems with [kmem arena] MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-BeenThere: freebsd-fs@freebsd.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: Filesystems List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 31 Jul 2015 09:46:21 -0000 Hello Freebsd-fs, I've rebuilt OS to ~latest version to have VM-related fixes, and still have problems with processes stuck in "kmem arena" state when they access files on ZFS with large blocks. I have ARC limited to 3GiB (out of 6GiB of physical memory), and ZFS with large files (1Gb+) and large blocks (16MiB) on radiz vdev (5 disks). Simple "dd if=large.file of=/dev/null bs=1m" could spent in "kmem arena" state Looks like such large blocks was bad idea, but anyway 500Kb/s linear reading? Really? What is strange, CPU is totally idle when reading process is stuck. No kernel threads are working for tens of minutes! I'm rebuilding pool with more resonable block sizes (and lets it waste space and time for metadata), but this situation (no work could be done when CPU is totally idle) looks wired and ugly anyway. -- Best regards, Lev mailto:lev@FreeBSD.org