cpio versus star

From: Dennis Clarke (dclarke_at_blastwave.org)
Date: 10/05/03

  • Next message: Robert Pouliot: "Re: Which PC-bucket RAM would work on an Ultra5?" 
    Date: Sun, 5 Oct 2003 10:08:05 -0700

    Normally when I want to move a directory structure from one machine to another
    I just use find and cpio thus :

    find . -depth -print | cpio -pdmV /destination

    Today I decided to give star a try with the following :

    star -c -Hexustar -acl -v . | star -xp -acl -C /destination

    In this case the destination is a NFS mount and the network is 100Mb full
    duplex with hme nics at each end.

    The sending server has netstats that look like so :

    # netstat -i -I hme0 5
        input hme0 output input (Total) output
    packets errs packets errs colls packets errs packets errs colls
    632475 0 1000674 0 0 632532 0 1000731 0 0
    909 0 922 0 0 909 0 922 0 0
    837 0 838 0 0 837 0 838 0 0
    844 0 843 0 0 844 0 843 0 0
    896 0 896 0 0 896 0 896 0 0
    910 0 909 0 0 910 0 909 0 0
    862 0 860 0 0 862 0 860 0 0
    920 0 928 0 0 920 0 928 0 0
    895 0 895 0 0 895 0 895 0 0
    909 0 920 0 0 909 0 920 0 0
    1120 0 1502 0 0 1120 0 1502 0 0
    893 0 1064 0 0 893 0 1064 0 0
    868 0 863 0 0 868 0 863 0 0
    942 0 941 0 0 942 0 941 0 0
    917 0 917 0 0 917 0 917 0 0
    863 0 863 0 0 863 0 863 0 0

    The receiving NFS server stats look like so :

    $ netstat -i -I hme0 5
        input hme0 output input (Total) output
    packets errs packets errs colls packets errs packets errs colls
    208501090 0 198926719 0 0 208501144 0 198926773 0 0
    941 0 942 0 0 941 0 942 0 0
    971 0 972 0 0 971 0 972 0 0
    938 0 938 0 0 938 0 938 0 0
    967 0 964 0 0 967 0 964 0 0
    921 0 921 0 0 921 0 921 0 0
    913 0 913 0 0 913 0 913 0 0
    948 0 945 0 0 948 0 945 0 0
    923 0 923 0 0 923 0 923 0 0
    914 0 914 0 0 914 0 914 0 0
    919 0 919 0 0 919 0 919 0 0
    1194 0 1054 0 0 1194 0 1054 0 0
    1071 0 957 0 0 1071 0 957 0 0

    This is really slow. I would expect much higher throughput from star than
    from cpio.

    The sending NFS client is on at 100Mb/sec full duplex thus :

    # /usr/sbin/ndd -set /dev/hme instance 0
    # ndd -get /dev/hme link_status
    1
    # ndd -get /dev/hme link_speed
    1
    # ndd -get /dev/hme link_mode
    1

    The NFS server is tha same.

    Did I miss something?

    Dennis

  • Next message: Robert Pouliot: "Re: Which PC-bucket RAM would work on an Ultra5?"

    Relevant Pages