Date: Mon, 7 Jan 2013 11:20:12 +0100 From: Fleuriot Damien <ml@my.gd> To: Daniel Kalchev <daniel@digsys.bg> Cc: Current FreeBSD <freebsd-current@freebsd.org>, "O. Hartmann" <ohartman@zedat.fu-berlin.de> Subject: Re: ZFS/RAIDZ and SAMBA: abyssimal performance Message-ID: <74D89D84-CC53-49A9-8D69-AF255A8323E0@my.gd> In-Reply-To: <A73352A5-12BE-4EC8-A5BC-C1D7C13E0A24@digsys.bg> References: <50E6DE91.7010404@zedat.fu-berlin.de> <1ADC2ECB-70FF-4DDD-9D62-16E2EEECDD8B@my.gd> <A73352A5-12BE-4EC8-A5BC-C1D7C13E0A24@digsys.bg>
next in thread | previous in thread | raw e-mail | index | archive | help
On Jan 7, 2013, at 11:14 AM, Daniel Kalchev <daniel@digsys.bg> wrote: >=20 > On Jan 4, 2013, at 4:06 PM, Fleuriot Damien <ml@my.gd> wrote: >=20 >>=20 >> And network cards: >> # Up a bit our intel cards parameters >> hw.em.txd=3D4096 >> hw.em.rxd=3D4096 >> hw.em.tx_int_delay=3D512 >> hw.em.rx_int_delay=3D512 >> hw.em.tx_abs_int_delay=3D1024 >> hw.em.rx_abs_int_delay=3D1024 >>=20 >=20 > I am curious why we need to manually set up these values. Especially = the txd/rxd -- here are few controllers supported by the em driver that = can't handle 4096 descriptors and the choice could really be made at = driver attach time.. That could also permit different em interfaces in = the system (using different chips) to have different settings. >=20 > My belief is the auto tuning should set things up for maximum = performance, given the hardware and if someone really needs smaller = queues they could just use the tunables.=20 >=20 > Are there drawbacks? >=20 > Daniel Well perhaps the code to handle auto tuning isn't present in the driver = itself. I'm not a huge fan of the idea, I believe it would be rather taxing to = implement all the exceptions and that some could easily be overlooked. I believe it's better to have a more user-friendly documentation and let = users tune the hardware to suit their needs.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?74D89D84-CC53-49A9-8D69-AF255A8323E0>