Professional Web Applications Themes

raw vs. ufs comparison - Sun Solaris

Hi there! someone wrote: > Did some quick tests of raw vs. ufs ... both under SDS 4.2.1 on Solaris8 > ... but it is turning out that ufs is faster. Does this seem reasonable ? It does. When using the raw device, any I/O is done synchronously. When using UFS, only metadata is written synchronously. For your "dd" example, the latter is surely more performant. The usual application of raw devices (or the "forcedirectio" flag) however is DBMSes, which do their own caching and perform any disk I/O with the O_SYNC flag. In that case raw devices win, because ...

  1. #1

    Default Re: raw vs. ufs comparison

    Hi there!

    someone wrote:
    > Did some quick tests of raw vs. ufs ... both under SDS 4.2.1 on Solaris8
    > ... but it is turning out that ufs is faster. Does this seem reasonable ?
    It does. When using the raw device, any I/O is done synchronously. When
    using UFS, only metadata is written synchronously. For your "dd" example,
    the latter is surely more performant.

    The usual application of raw devices (or the "forcedirectio" flag) however
    is DBMSes, which do their own caching and perform any disk I/O with the
    O_SYNC flag. In that case raw devices win, because data doesn't need to be
    copied from one memory region to another (from application memory to cache),
    and is instead written directly from user space memory to disk.

    The OS'es caching mechanisms mainly just increase I/O overhead if you're
    running a DBMS.

    --
    Jan Andres
    Jan Andres Guest

  2. #2

    Default Re: raw vs. ufs comparison

    Jan Andres wrote:
    > > Did some quick tests of raw vs. ufs ... both under SDS 4.2.1 on Solaris8
    > > ... but it is turning out that ufs is faster. Does this seem reasonable ?
    >
    > It does. When using the raw device, any I/O is done synchronously.
    Is that right? Guess the aiowrite man page must be wrong then.
    > The usual application of raw devices (or the "forcedirectio" flag) however
    > is DBMSes, which do their own caching and perform any disk I/O with the
    > O_SYNC flag.
    The O_SYNC flag with raw devices. Amazing!
    > In that case raw devices win, because data doesn't need to be
    > copied from one memory region to another (from application memory to cache),
    > and is instead written directly from user space memory to disk.
    And that makes it faster? I didn't realise memory copies are
    so slow.

    -am 2003
    Anthony Mandic Guest

  3. #3

    Default Re: raw vs. ufs comparison

    noone <noonenoone.org> wrote:
    > Did some quick tests of raw vs. ufs ... both under SDS 4.2.1 on Solaris8
    > ... but it is turning out that ufs is faster. Does this seem reasonable ?
    Have you installed the latest SDS 4.2.1 patch (108693 on SPARC)? I
    remember, the SDS 4.2.1 without this patch had some serious
    performance problems when accessing the raw metadevices under certain
    configurations.


    -akop
    Akop Pogosian Guest

  4. #4

    Default Re: raw vs. ufs comparison

    In article <beun6d$rek$1lust.ihug.co.nz>,
    jmsalvo <jmsalvo.au> writes:
    >
    > A much better tool to do the raw vs. ufs should be dt ( similar to dd ),
    > with support for async i/o:
    >
    > [url]http://www.bit-net.com/~rmiller/dt.html[/url]
    >
    > However, the PDF manual says that aio support in the program is only
    > built-in for Tru64.
    >
    > Anyone tried it on Solaris ?
    Along those lines, there's also
    [url]ftp://ftp.lysator.liu.se/pub/unix/pcopy/pcopy-1.3.tar.gz[/url], a threaded
    copy program (although I suppose asyncio could be faster than threads,
    pthreads are arguably more widely available). It might be interesting
    if someone with large enough amounts of disk space to for significant
    results timed some of the various copy programs out there.

    --
    mailto:rlhamilmindwarp.smart.net [url]http://www.smart.net/~rlhamil[/url]
    Richard L. Hamilton Guest

  5. #5

    Default Re: raw vs. ufs comparison

    Richard L. Hamilton wrote:
    > In article <beun6d$rek$1lust.ihug.co.nz>,
    > jmsalvo <jmsalvo.au> writes:
    >
    >>A much better tool to do the raw vs. ufs should be dt ( similar to dd ),
    >>with support for async i/o:
    >>
    >>[url]http://www.bit-net.com/~rmiller/dt.html[/url]
    >>
    >>However, the PDF manual says that aio support in the program is only
    >>built-in for Tru64.
    >>
    >>Anyone tried it on Solaris ?
    >
    >
    > Along those lines, there's also
    > [url]ftp://ftp.lysator.liu.se/pub/unix/pcopy/pcopy-1.3.tar.gz[/url], a threaded
    > copy program (although I suppose asyncio could be faster than threads,
    > pthreads are arguably more widely available). It might be interesting
    > if someone with large enough amounts of disk space to for significant
    > results timed some of the various copy programs out there.
    >
    Interesting ... but I really wanted to test async I/O on raw vs. async
    I/O with concurrent direct I/O on UFS.

    The dt program from Robert Miller that I mentioned above does not link
    against the aio library.


    John Salvo Guest

  6. #6

    Default Re: raw vs. ufs comparison

    In article <Am0Ra.5927$wU5.3080news-server.bigpond.net.au>,
    John Salvo <jmsalvo.au> writes:
    > Richard L. Hamilton wrote:
    >> In article <beun6d$rek$1lust.ihug.co.nz>,
    >> jmsalvo <jmsalvo.au> writes:
    >>
    >>>A much better tool to do the raw vs. ufs should be dt ( similar to dd ),
    >>>with support for async i/o:
    >>>
    >>>[url]http://www.bit-net.com/~rmiller/dt.html[/url]
    >>>
    >>>However, the PDF manual says that aio support in the program is only
    >>>built-in for Tru64.
    >>>
    >>>Anyone tried it on Solaris ?
    >>
    [...]
    > The dt program from Robert Miller that I mentioned above does not link
    > against the aio library.
    Well, it wouldn't directly, even if it were set up to do AIO on Solaris,
    it would link against -lrt, which would pull in libaio because the librt
    POSIX aio functions are implemented in terms of libaio on Solaris.

    If you add dtaio.c to CFILES, -lrt to EXTLIBS, and -DAIO to CFLAGS
    in Makefile.solaris, it should compile with the AIO functionality enabled
    on Solaris 8 (it did for me). No idea whether it will actually _work_ or
    not, I leave that up to you. :-)

    --
    mailto:rlhamilmindwarp.smart.net [url]http://www.smart.net/~rlhamil[/url]
    Richard L. Hamilton Guest

Similar Threads

  1. Comparison among different PDF libraries
    By deepak_goyal@adobeforums.com in forum Adobe Acrobat SDK
    Replies: 6
    Last Post: August 19th, 02:34 PM
  2. Easy for u--comparison
    By emmim44 in forum Coldfusion - Advanced Techniques
    Replies: 1
    Last Post: July 10th, 04:21 PM
  3. Comparison of unices
    By Ed in forum Linux / Unix Administration
    Replies: 4
    Last Post: January 19th, 10:40 PM
  4. ASP comparison question...
    By K in forum ASP
    Replies: 3
    Last Post: August 1st, 03:39 PM
  5. Mac - PC Comparison
    By Ed A. Ortiz in forum Adobe Photoshop 7, CS, CS2 & CS3
    Replies: 6
    Last Post: July 16th, 07:49 PM

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139