From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jason L Tibbitts III Subject: Re: Sporadic timeouts accessing shingled drives with 3.19. Date: Mon, 20 Apr 2015 18:01:38 -0500 Message-ID: References: Mime-Version: 1.0 Content-Type: text/plain Return-path: Received: from mx2.math.uh.edu ([129.7.128.33]:40051 "EHLO mx2.math.uh.edu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751350AbbDTXBk (ORCPT ); Mon, 20 Apr 2015 19:01:40 -0400 In-Reply-To: (Adrian Palmer's message of "Mon, 20 Apr 2015 15:36:40 -0600") Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Adrian Palmer Cc: linux-ide@vger.kernel.org >>>>> "AP" == Adrian Palmer writes: AP> I'm the Seagate engineer leading SMR/ZAC/ZBC support in the kernel. AP> I would be very interested in learning about your setup and AP> workload. There's not much to it, really; I just needed a lot of space. The machine hasn't really been doing all that much yet besides RAID resyncs. Its function is a security system. IP cameras connect to a web server, push a video file which gets saved in RAM. Once done, that file gets converted to MP4 by ffmpeg with the output going to disk. A smaller thumbnail file (lower resolution and sped up) is also generated and written to disk. Then an sqlite file is updated and a log file is written. That's... about it. Total writes are about 250GB/day, which happens to be less than the amount of RAM in the system. So the bulk of the writes are streaming, and no temporary files ever hit the disk. Of course, the OS is there doing whatever writes it feels it needs to do. AP> There are 3 types of SMR drives: Drive Managed (DM), Host Managed AP> (HM), and Host Aware (HA). I suspect that you have DM drives. Definitely. These are just the plain 8TB "archive" disks that Seagate offers, purchased from Newegg. Nothing fancy. The model info was in the logs: ST8000AS0002-1NA17Z, AR13. AP> With DM, there is both nothing done, and nothing to be done in the AP> kernel. The mechanics of SMR are implemented in the drive, AP> completely abstracted from the system. The use case is not targeted AP> toward RAID applications -- as of yet, I've seen poor performance AP> from around the web from those that have attempted it. Well, performance isn't really the goal here. I'm not sure that "RAID applications" has any general meaning; if you mean heavy use file server stuff, sure, the drives are terrible for that. But the fact that the drives are in an array doesn't imply any usage pattern at all. I just needed a lot of space and some basic redundancy. However, regardless of the use case at which the drives are targeted, they should at least not drop off the bus. Now, at this point I can't tell if the fault lies in the drives, but given that I'm cranking along just fine on 3.18.9 while 3.19.4 is mostly useless I know where I'd look first. - J<