Re: fxp performance with POLLING - FreeBSD

This is a discussion on Re: fxp performance with POLLING - FreeBSD ; * Bartosz Stec [081003 07:23] wrote: > Hello again > > With POLLING enabled I experience about 10%-25% performance drop when > copying files over network. Tested with both SAMBA and NFS. Is it normal? > > FreeBSD 7.1-PRERELEASE #0: ...

+ Reply to Thread
Results 1 to 9 of 9

Thread: Re: fxp performance with POLLING

  1. Re: fxp performance with POLLING

    * Bartosz Stec [081003 07:23] wrote:
    > Hello again
    >
    > With POLLING enabled I experience about 10%-25% performance drop when
    > copying files over network. Tested with both SAMBA and NFS. Is it normal?
    >
    > FreeBSD 7.1-PRERELEASE #0: Sat Sep 6 01:52:12 CEST 2008
    > fxp0: port 0xc800-0xc83f mem
    > 0xe1021000-0xe1021fff irq 20 at device 8.0 on pci1
    >
    > # ifconfig fxp0
    > fxp0: flags=9843
    > metric 0 mtu 1500
    > options=8
    > ether 00:20:ed:42:87:13
    > inet 192.168.0.2 netmask 0xffffff00 broadcast 192.168.0.255
    > media: Ethernet autoselect (100baseTX )
    > status: active
    >
    > BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    > RELENG_5 ... - 7.5 MB/s peak.


    7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    not "sucks".

    Where do you see faster performance?

    Between windows machines on the same hardware or linux server?

    --
    - Alfred Perlstein
    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  2. Re: fxp performance with POLLING

    Alfred Perlstein wrote:
    > * Bartosz Stec [081003 07:23] wrote:
    >
    >> Hello again
    >>
    >> With POLLING enabled I experience about 10%-25% performance drop when
    >> copying files over network. Tested with both SAMBA and NFS. Is it normal?
    >>
    >> FreeBSD 7.1-PRERELEASE #0: Sat Sep 6 01:52:12 CEST 2008
    >> fxp0: port 0xc800-0xc83f mem
    >> 0xe1021000-0xe1021fff irq 20 at device 8.0 on pci1
    >>
    >> # ifconfig fxp0
    >> fxp0: flags=9843
    >> metric 0 mtu 1500
    >> options=8
    >> ether 00:20:ed:42:87:13
    >> inet 192.168.0.2 netmask 0xffffff00 broadcast 192.168.0.255
    >> media: Ethernet autoselect (100baseTX )
    >> status: active
    >>
    >> BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    >> RELENG_5 ... - 7.5 MB/s peak.
    >>

    >
    > 7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    > not "sucks".
    >
    > Where do you see faster performance?
    >
    > Between windows machines on the same hardware or linux server?
    >
    >

    It sucks because it is a peak performance. About 5-6 MB/s average. I
    tried polling only because I found some suggestions on mailing lists,
    that it could improve performance with SAMBA on FreeBSD. As you see at
    the top of this thread - not in my case I also tried sysctl tunings,
    and smb.conf settings, also suggested on maling lists, with no or very
    little improvements noticed. Most of suggestions unfortunately end with
    "change OS to Linux if you want to use SAMBA". I think I will try to
    change NIC to 1Gbit - hope that helps Or maybe there's some "FreeBSD
    and SAMBA tuning guide" which I didn't found?

    --
    Bartosz Stec

    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  3. Re: fxp performance with POLLING

    On Mon, Oct 06, 2008 at 09:02:33AM +0200, Bartosz Stec wrote:
    > Alfred Perlstein wrote:
    >> * Bartosz Stec [081003 07:23] wrote:
    >>
    >>> Hello again
    >>>
    >>> With POLLING enabled I experience about 10%-25% performance drop when
    >>> copying files over network. Tested with both SAMBA and NFS. Is it
    >>> normal?
    >>>
    >>> FreeBSD 7.1-PRERELEASE #0: Sat Sep 6 01:52:12 CEST 2008
    >>> fxp0: port 0xc800-0xc83f mem
    >>> 0xe1021000-0xe1021fff irq 20 at device 8.0 on pci1
    >>>
    >>> # ifconfig fxp0
    >>> fxp0: flags=9843
    >>> metric 0 mtu 1500
    >>> options=8
    >>> ether 00:20:ed:42:87:13
    >>> inet 192.168.0.2 netmask 0xffffff00 broadcast 192.168.0.255
    >>> media: Ethernet autoselect (100baseTX )
    >>> status: active
    >>>
    >>> BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    >>> RELENG_5 ... - 7.5 MB/s peak.
    >>>

    >>
    >> 7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    >> not "sucks".
    >>
    >> Where do you see faster performance?
    >>
    >> Between windows machines on the same hardware or linux server?
    >>
    >>

    > It sucks because it is a peak performance. About 5-6 MB/s average. I
    > tried polling only because I found some suggestions on mailing lists,
    > that it could improve performance with SAMBA on FreeBSD. As you see at
    > the top of this thread - not in my case I also tried sysctl tunings,
    > and smb.conf settings, also suggested on maling lists, with no or very
    > little improvements noticed. Most of suggestions unfortunately end with
    > "change OS to Linux if you want to use SAMBA". I think I will try to
    > change NIC to 1Gbit - hope that helps Or maybe there's some "FreeBSD
    > and SAMBA tuning guide" which I didn't found?


    Can you please test network I/O using something like netperf or one of
    the other network-benchmark tools and not things like NFS or Samba
    which rely on disk I/O and other aspects?

    --
    | Jeremy Chadwick jdc at parodius.com |
    | Parodius Networking http://www.parodius.com/ |
    | UNIX Systems Administrator Mountain View, CA, USA |
    | Making life hard for others since 1977. PGP: 4BD6C0CB |

    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  4. Re: fxp performance with POLLING

    Jeremy Chadwick wrote:
    > On Mon, Oct 06, 2008 at 09:02:33AM +0200, Bartosz Stec wrote:
    >
    >> Alfred Perlstein wrote:
    >>
    >>> * Bartosz Stec [081003 07:23] wrote:
    >>>
    >>>
    >>>> Hello again
    >>>>
    >>>> With POLLING enabled I experience about 10%-25% performance drop when
    >>>> copying files over network. Tested with both SAMBA and NFS. Is it
    >>>> normal?
    >>>>
    >>>> FreeBSD 7.1-PRERELEASE #0: Sat Sep 6 01:52:12 CEST 2008
    >>>> fxp0: port 0xc800-0xc83f mem
    >>>> 0xe1021000-0xe1021fff irq 20 at device 8.0 on pci1
    >>>>
    >>>> # ifconfig fxp0
    >>>> fxp0: flags=9843
    >>>> metric 0 mtu 1500
    >>>> options=8
    >>>> ether 00:20:ed:42:87:13
    >>>> inet 192.168.0.2 netmask 0xffffff00 broadcast 192.168.0.255
    >>>> media: Ethernet autoselect (100baseTX )
    >>>> status: active
    >>>>
    >>>> BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    >>>> RELENG_5 ... - 7.5 MB/s peak.
    >>>>
    >>>>
    >>> 7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    >>> not "sucks".
    >>>
    >>> Where do you see faster performance?
    >>>
    >>> Between windows machines on the same hardware or linux server?
    >>>
    >>>
    >>>

    >> It sucks because it is a peak performance. About 5-6 MB/s average. I
    >> tried polling only because I found some suggestions on mailing lists,
    >> that it could improve performance with SAMBA on FreeBSD. As you see at
    >> the top of this thread - not in my case I also tried sysctl tunings,
    >> and smb.conf settings, also suggested on maling lists, with no or very
    >> little improvements noticed. Most of suggestions unfortunately end with
    >> "change OS to Linux if you want to use SAMBA". I think I will try to
    >> change NIC to 1Gbit - hope that helps Or maybe there's some "FreeBSD
    >> and SAMBA tuning guide" which I didn't found?
    >>

    >
    > Can you please test network I/O using something like netperf or one of
    > the other network-benchmark tools and not things like NFS or Samba
    > which rely on disk I/O and other aspects?
    >
    >

    I remember when on FreeBSD 4.x I was able to copy files from samba and
    to samba up to 12MB/s on 100Mbit lan. Now with FreeBSD 6.x or 7.x I can
    have barely 8MB/s on 100Mbit and 27MB/s on Gigabit lan. Netperf shows
    900Mbit/s in any direction, small variety with different switches (ca
    10% difference).
    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  5. Re: fxp performance with POLLING

    Bartosz Stec wrote:

    >>> BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    >>> RELENG_5 ... - 7.5 MB/s peak.
    >>>

    >>
    >> 7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    >> not "sucks".
    >>
    >> Where do you see faster performance?
    >>
    >> Between windows machines on the same hardware or linux server?
    >>
    >>

    > It sucks because it is a peak performance. About 5-6 MB/s average. I
    > tried polling only because I found some suggestions on mailing lists,
    > that it could improve performance with SAMBA on FreeBSD. As you see at
    > the top of this thread - not in my case I also tried sysctl tunings,
    > and smb.conf settings, also suggested on maling lists, with no or very
    > little improvements noticed. Most of suggestions unfortunately end with
    > "change OS to Linux if you want to use SAMBA". I think I will try to
    > change NIC to 1Gbit - hope that helps Or maybe there's some "FreeBSD
    > and SAMBA tuning guide" which I didn't found?
    >


    Please try experimenting with "socket options" in smb.conf, I've found
    that some tuning is desirable on any OS with Samba, but these are the
    values that worked best for me with Windows XP clients in mind. Win2003
    clients seemed much faster without tuning (same base code as XP 64bit)
    and I suspect it has a different SMB implementation. I'd suggest
    starting with "socket options = TCP_NODELAY IPTOS_THROUGHPUT
    SO_RCVBUF=8192 SO_SNDBUF=8192" and if you aren't satisfied, experiment
    with the numbers and which options are enabled. Be sure that the
    client has been disconnected from Samba completely to make sure you are
    testing the values in the config file. I'm pretty sure with these
    tunings I was able to get closer to 10MB/sec on 100Mbit, which satisfies
    me for the average user.

    # Most people will find that this option gives better performance.
    # See smb.conf(5) and /usr/share/doc/samba-doc/htmldocs/speed.html
    # for details
    # You may want to add the following on a Linux system:
    # SO_RCVBUF=8192 SO_SNDBUF=8192
    # socket options = TCP_NODELAY
    # For some reason, 8192 is pretty fast on a XP lab 100Mb client. Other
    sizes tested and dissapointing in that situation. Windows Server 2k3 on
    gig is much faster, and likes larger values. There might be some merit
    in testing 49152 in some situations. (20080617)
    # TCP_NODELAY makes a huge improvement. IPTOS_THROUGHPUT is negligible
    locally.
    # mcdouga9 20070110
    socket options = TCP_NODELAY IPTOS_THROUGHPUT SO_RCVBUF=8192
    SO_SNDBUF=8192
    # socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192


    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  6. Re: fxp performance with POLLING

    Clifton Royston wrote:
    > On Mon, Oct 06, 2008 at 10:29:35AM +0300, Andrei Kolu wrote:
    > ...
    >
    >> I remember when on FreeBSD 4.x I was able to copy files from samba and
    >> to samba up to 12MB/s on 100Mbit lan.
    >>

    >
    > This part seems unlikely, particularly as bit rates are measured in
    > decimal millions not computer millions.
    >
    > 12*8*1024*1024 = 100,663,296 so that would mean not merely zero but
    > negative packet and network overhead.
    >
    > -- Clifton
    >

    OK, I am reading right now description for Trendnet Mini-GBIC Features:
    --------------------------------------------------------------------
    (TEG-MGBSX, TEG-MGBS10, TEW-MGBS40, TEW-MGBS80)
    Compliant with IEEE 802.3z Gigabit Ethernet and Fiber Channel Standards
    Industry standard SFP package
    Duplex LC connector
    1.0625Gbps Fiber Channel Compliant
    1.25Gbps Gigabit Ethernet Compliant
    --------------------------------------------------------------------
    So, I guess that 100Mbit and 1000Mbit is not set in stone and you can
    actually achieve higher speeds than "standard".

    BTW:

    1 megabit = 106 = 1,000,000 bits which is equal to 125,000 bytes.
    100 megabit = 12,500,000 bytes = 12,5MB

    Or I am wrong?
    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  7. Re: fxp performance with POLLING

    > 1 megabit = 106 = 1,000,000 bits which is equal to 125,000 bytes.

    you are assuming eight bits per byte - but this is a serial line so
    you should use ten bits per byte instead.

    -pete.

    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  8. Re: fxp performance with POLLING

    On Mon, Oct 06, 2008 at 08:38:13AM -0400, Adam McDougall wrote:
    > Bartosz Stec wrote:
    >
    >>>> BTW overall SAMBA performance still sucks on 7.1-pre as much as on
    >>>> RELENG_5 ... - 7.5 MB/s peak.
    >>>>
    >>>
    >>> 7.5MB is 75% effeciency of a 100mbit card. Not amazing, but
    >>> not "sucks".
    >>>
    >>> Where do you see faster performance?
    >>>
    >>> Between windows machines on the same hardware or linux server?
    >>>
    >>>

    >> It sucks because it is a peak performance. About 5-6 MB/s average. I
    >> tried polling only because I found some suggestions on mailing lists,
    >> that it could improve performance with SAMBA on FreeBSD. As you see at
    >> the top of this thread - not in my case I also tried sysctl tunings,
    >> and smb.conf settings, also suggested on maling lists, with no or very
    >> little improvements noticed. Most of suggestions unfortunately end with
    >> "change OS to Linux if you want to use SAMBA". I think I will try to
    >> change NIC to 1Gbit - hope that helps Or maybe there's some "FreeBSD
    >> and SAMBA tuning guide" which I didn't found?
    >>

    >
    > Please try experimenting with "socket options" in smb.conf, I've found
    > that some tuning is desirable on any OS with Samba, but these are the
    > values that worked best for me with Windows XP clients in mind. Win2003
    > clients seemed much faster without tuning (same base code as XP 64bit)
    > and I suspect it has a different SMB implementation. I'd suggest
    > starting with "socket options = TCP_NODELAY IPTOS_THROUGHPUT
    > SO_RCVBUF=8192 SO_SNDBUF=8192" and if you aren't satisfied, experiment
    > with the numbers and which options are enabled. Be sure that the
    > client has been disconnected from Samba completely to make sure you are
    > testing the values in the config file. I'm pretty sure with these
    > tunings I was able to get closer to 10MB/sec on 100Mbit, which satisfies
    > me for the average user.


    Interesting. I use gigE on my home network, and most of my Samba shares
    consist of very large files (ISO images and HD videos). Setting the
    buffer sizes to 8KB results in horrible performance; a 420MB ISO image
    taking ~50 seconds to copy. Copying is from the Samba server to the
    Windows machine. ZFS is used on FreeBSD, across 4 disks in a raidz1
    pool; all disks are SATA300 with 16MByte cache.

    Increasing the Samba buffers to 64KB (65536) drops the transfer to 30
    seconds, and increasing it to 128KB (131072) drops to 15-20
    seconds.

    NICs involved are an Attansic L1E (on Windows XP, with TCP buffer sizes
    tuned in the registry, ditto with enabling TCP window resizing), and an
    em(4) (on FreeBSD). There are two switches between the boxes: an HP
    ProCurve, and a generic D-Link switch. Everything is gigE.

    Additionally, don't forget about IPTOS_LOWDELAY, and the "write cache
    size" and "read size" options in smb.conf.

    --
    | Jeremy Chadwick jdc at parodius.com |
    | Parodius Networking http://www.parodius.com/ |
    | UNIX Systems Administrator Mountain View, CA, USA |
    | Making life hard for others since 1977. PGP: 4BD6C0CB |

    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


  9. Re: fxp performance with POLLING

    Pete French wrote:
    >> 1 megabit = 106 = 1,000,000 bits which is equal to 125,000 bytes.
    >>

    >
    > you are assuming eight bits per byte - but this is a serial line so
    > you should use ten bits per byte instead.
    >
    > -pete.
    >


    That was a rule of thumb in the heyday of async serial lines, which used
    a start and stop bit per byte.

    However, ethernet at 100Mbit is 4B5B coded at a 125mhz rate. So the raw
    synchronous data rate really is 12.5Mbytes/s. Minus the sync preamble
    of 8 bytes per packet and the mandatory inter-frame-gap of 12 bytes
    that's a physical layer rate of (12.5M * (1500/(1500+20))) or 12.34Mbyte/s.

    Even in the later days of modems this rule applied less and less,
    because the modulation schemes became synchronous.

    Joe Koberg
    joe_at_osoft_dot_us


    _______________________________________________
    freebsd-stable@freebsd.org mailing list
    http://lists.freebsd.org/mailman/lis...freebsd-stable
    To unsubscribe, send any mail to "freebsd-stable-unsubscribe@freebsd.org"


+ Reply to Thread