Re: cp taking 24 hours for 200Gigs of data

From: Logan Shaw (lshaw-usenet_at_austin.rr.com)
Date: 03/31/04 

Date: Wed, 31 Mar 2004 07:36:40 GMT

Liam Carey wrote:

> We are having to wait over 24 hours for cp to complete the job of
> copying 200Gig of data from one filesystem to another. Any help is
> greatly appreciated. Both machines are Solaris 8, with Gigabit
> ethernet cards. We have mounted the filesytem via NFS over the network
> and are currently copying 3 directories we need using cp -r.

First recommendation is to turn on the "logging" mount option for
the destination filesystem if you haven't done that.

Second recommendation is to not use NFS for bulk copying. For
small files, you will always have to contend with the fact that
there is turnaround time for every file that is created, and that
turnaround time is going to be increased by not only the latency
of the network (which is low) but also the latency of the software
answering a queue of network requests. And yes, you can have
multiple requests in progress at once and reduce the effect of
this turnaround time, but since this is (probably) the one
factor that dominates, that is just going to reduce the slowness
rather than eliminating it.

I personally would do something like this:

        ufsdump 0f - /source/filesystem |
             rsh destination-host "cd /dest/dir && ufsrestore rf -"

It is longer and more work, but I would almost bet money it
will go a LOT faster.

Of course Joerg will probably pop up and point out that the
performance will be even better with "star", which is
probably true, but you may find that ufsdump is adequate
for your needs.

   - Logan

Relevant Pages

  • Re: OT: why no file copy() libc/syscall ??
    ... > Now if you wanted the remote server to deny the network copy... ... > support that may be available for copying files. ... If you really want a filesystem that supports efficient copying you ...
    (Linux-Kernel)
  • [0/5] New POHMELFS release.
    ... This is a new release of the POHMEL filesystem. ... POHMELFS stands for Parallel Optimized Host Message Exchange Layered File System. ... Network filesystem is a client transport. ... Very fast and scalable multithreaded userspace server. ...
    (Linux-Kernel)
  • Re: Security and Sharing
    ... When they are logged in locally only the filesystem permissions are needed. ... When they access over the network they can do anything that the filesystem ... If you want then to be able to read files and browse the folder structure ...
    (microsoft.public.security)
  • Re: [ANNOUNCEMENT PATCH COW] proof of concept impementation of cowlinks
    ... > are involved) and makes error handling and recovery for network ... > filesystems much harder since inconsistent client and server state have ... > filesystem protocol takes in effect four parameters (there is no handle ... send the line "unsubscribe linux-kernel" in ...
    (Linux-Kernel)
  • Re: [ANNOUNCEMENT PATCH COW] proof of concept impementation of cowlinks
    ... SMB/CIFS - there is no equivalent network protocol operation. ... roundtrip delay - open/sendfile/close sends at a minimum three times as ... filesystem protocol takes in effect four parameters (there is no handle ... And the shares referenced have to be on the same server ...
    (Linux-Kernel)