Date: Mon, 15 Feb 2016 12:33:04 -0800 From: Ravi Pokala <rpokala@mac.com> To: "freebsd-fs@freebsd.org" <freebsd-fs@freebsd.org> Subject: Re: Hours of tiny transfers at the end of a ZFS resilver? Message-ID: <8E04E52A-2635-4253-8140-F69495D7D0A6@panasas.com>
next in thread | raw e-mail | index | archive | help
>Date: Mon, 15 Feb 2016 21:18:59 +1100 >From: Andrew Reilly <areilly@bigpond.net.au> >To: freebsd-fs@freebsd.org >Subject: Hours of tiny transfers at the end of a ZFS resilver? >Message-ID: <120226C8-3003-4334-9F5F-882CCB0D28C5@bigpond.net.au> >Content-Type: text/plain; charset=us-ascii > >Hi Filesystem experts, Hi Andrew, I am in no way, shape, or form a filesystem expert. :-) I *am*, however, an ATA drive expert. I wanted to clarify something you said, because it seems to be a common misunderstanding. >... the new drives actually have 4096B sectors (although they lie about that in smartctl -i queries): They're not lying. >Sector Sizes: 512 bytes logical, 4096 bytes physical Right there - "Sector Size: ... 4096 bytes physical". This is 512B logical / 4KB physical scheme is called AF-512e, and is a documented, standard format. https://en.wikipedia.org/wiki/Advanced_Format#512e The intent is to allow backwards-compatibility with software going back decades, which only knows about 512-byte sectors. Such software would treat an AF-512e drive the same as any other drive. The trade-off is performance, because the drive has to transparently perform read-modify-write operations (for sub-4096B writes), and read the full 4096B physical sector even if only a single 512B logical sector was requested. (Ditto if a properly-sized but un-aligned request were made.) I know GEOM reports both the logical and physical sector sizes (as the provider's "sectorsize" and "stripesize", respectively), and I know that the ATACAM driver is populating them correctly based on the drive's IDENTIFY_DEVICE information. (My very first submission to the project was to fix some bugs in this code; r262886, 2014-03-07.) [*] I *don't* know if ZFS does the right thing automatically; it might not be able to determine what "the right thing" is in all cases. I leave answering that to the actual ZFS experts. :-) -Ravi (rpokala@) [*] This is probably a good segue into discussing why we even have the ADA_Q_4K quirk, and whether we should get rid of it...? --rp
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?8E04E52A-2635-4253-8140-F69495D7D0A6>