ZFS-lighttpd2-sendfile, too high IO

[K. Macy]

2011/9/12 Steven Hartland <killing at multiplay.co.uk>:
> sendfile doesn't work as you might expect on zfs, its not zero copy due to
> the use of none buffer pool memory.
>
> You do get a benefit but it requires double the amount of memory to get it
> so we've disabled sendfile
> under nginx for zfs based hosts for this very reason.


In my performance testing, sending from mmaped I/O on ZFS was
dramatically faster (> 2x) than sendfile.

Cheers

>   Regards
>   Steve
> ----- Original Message ----- From: "Laszlo KAROLYI" <szoftos at freemail.hu>
> To: <freebsd-fs at freebsd.org>
> Sent: Monday, September 12, 2011 12:10 PM
> Subject: ZFS-lighttpd2-sendfile, too high IO
>
>
>> Hello, Recently I installed a FreeBSD with the newest 8.2-STABLE and zfs
>> version. I use lighttpd2 and zfs on it, and munin to monitor the outgoing
>> bandwidth. Zpool version 28, zfs version 5, with the latest kernel. We have
>> a big mp3 archive (half hour, 256kbit/s mp3-s) which lighty serves. This
>> means full random IO. When I see that the server serves 15mbit/sec, i see
>> constantly 8-10Mbyte/sec reads on the zfs raidz1 array, which is too much.
>> As I could see from truss logs, lighty uses sendfile and writev.My settings:
>> vfs.zfs.l2c_only_size: 15106737664
>
> ...
>
> ================================================
> This e.mail is private and confidential between Multiplay (UK) Ltd. and the
> person or entity to whom it is addressed. In the event of misdirection, the
> recipient is prohibited from using, copying, printing or otherwise
> disseminating it or any information contained in it.
> In the event of misdirection, illegible or incomplete transmission please
> telephone +44 845 868 1337
> or return the E.mail to postmaster at multiplay.co.uk.
>
> _______________________________________________
> freebsd-fs at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-fs
> To unsubscribe, send any mail to "freebsd-fs-unsubscribe at freebsd.org"
>