5 Replies Latest reply on Jul 1, 2015 3:13 PM by Bill Robinson

    Is there any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...?

    Pedro José Barbero Iglesias

      Hi there,

       

      I would like to know if there is any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...

       

      I found how this command "nshopt" could help us to find the better configuration for it... but I don't understand well how it works, due to the lack of documentation about it.

       

      I need this because the server we were using to access some NSH Scripts and other files used by our automatisms has started to be overwhelmed,  after moving the File-system located in it to a NFS volume and therefore multiplying the I/O by 2. 

       

      My best regards.

        • 2. Re: Is there any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...?
          Pedro José Barbero Iglesias

          Hi Bill,

           

          No I didn't, I am sorry I only typed "shot --help"... anyway, according to it. Should I either set the size of the write buffer at the server where NSH Console is,NFS volumen is mounted and the one being referenced by the Apps Servers, on their calls or should I set this write buffer's size at both, App Servers and NSH Console where the NTFS volume is mounted? And where should I set this, at the .nshrc file?

           

          I suppose this should be set at server where the volumen NFS is mounted and the one referenced by the invocations from the Apps Servers.

           

          Regards.

           

          ---------------------------------

           

          NSHOPT

                 nshopt - Test different network write buffer sizes

           

          SYNOPSIS

                 nshopt [-i size] [-k size] [-s bytes] [-b] host1 ...

           

          DESCRIPTION

                 Depending on the network, using specific write buffer sizes when communicating with remote hosts can

                 improve the net throughput of data. The default write buffer size is 4480 bytes, but sometimes  this

                 value may not be optimal.

           

                 To  determine  the  optimal  write  buffer  size, nshopt writes a 2MB file to a remote host multiple

                 times, each time using different network write buffer sizes and determining the  time  it  takes  to

                 send the file. This lets you determine the optimal network write buffer size to use when communicat-

                 ing with the given host.

           

                 nshopt starts with a write buffer size of 512 bytes and continues to perform the test  in  512  byte

                 increments  up  to  a  maximum buffer size of 16384 bytes (16KB).  nshopt prints the results of each

                 test to the standard output for review. (See EXAMPLE.)

           

                 Once nshopt has determined an optimal buffer size, use the secadmin command  to  configure  the  new

                 buffer size.

           

          OPTIONS

                 -i size Instead  of  starting  with  a write buffer size of 512 and using an increment of 512 bytes,

                         start with a write buffer size and use an increment size of size.

           

                 -k size Instead of transferring a 2 MB (2048 KB) test file as a sample, use a file size KB large.

           

                 -s bytes

                         Start off with a buffer size of bytes.  By default nshopt starts with a buffer size  equiva-

                         lent to the increment size (512 bytes).

           

                 -b      When  writing data to the remote host, perform a bulk write rather than a regular write. The

                         difference between the two is that with a bulk write there is no checking or return code  to

                         verify  that the write actually worked. A regular write does perform those checks and there-

                         fore will take a little longer. The cp command performs bulk writes when copying a file to a

                         remote host.

           

          EXAMPLE

                 The  following  example  tests  the  host hpux. From the data you can see that a buffer size of 1024

                 bytes is optimal for transferring data from the local host to the host hpux.  This example then uses

                 the command secadmin to update the configuration file with the desired buffer size.

           

                        # nshopt hpux

                        Trying   512 bytes to hpux ... done. (52.012 seconds for 2048 KB = 39 KB/sec)

                        Trying  1024 bytes to hpux ... done. (3.020 seconds for 2048 KB = 678 KB/sec)

                        Trying  1536 bytes to hpux ... done. (51.173 seconds for 2048 KB = 40 KB/sec)

                        Trying  2048 bytes to hpux ... done. (51.145 seconds for 2048 KB = 40 KB/sec)

                        Trying  2560 bytes to hpux ... done. (51.147 seconds for 2048 KB = 40 KB/sec)

                        Trying  3072 bytes to hpux ...

                         .

                         .

                        # secadmin -W hpux 1024

           

          CAVEATS

                 The  nshopt  command  tests how best to send data to a remote host. It does not test how fast it can

                 receive data.  If you anticipate that you will be receiving large amounts of data, then  you  should

                 be  running  this test from the agent server to the client server (where you will need to install an

                 agent to test it properly).

           

          ORIGIN

                 nshopt was written by Thomas Kraus.

           

          SEE ALSO

                 secadmin(1), secure(1), cp(1).

           

          --------------------------

          • 3. Re: Is there any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...?
            Bill Robinson

            it's all there in the man page:

             

                   Once nshopt has determined an optimal buffer size, use the secadmin command  to  configure  the  new

                   buffer size.


            The  following  example  tests  the  host hpux. From the data you can see that a buffer size of 1024

            ...

            secadmin -W hpux 1024



            • 4. Re: Is there any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...?
              Pedro José Barbero Iglesias

              Hi again Bill,

               

              Ok I see... so that,I determined that the place where I should set the new buffer configuration at the server from which the files are being retrieved. Because is the one who is going to write them through network interfece as response...

               

              The way these files are requested from the App Servers is ... "nsh -c "//file_server_fqdn/bladelogic/bsa/NSH/storage/extended_objects/script_name.nsh ??TARGET.HOST??""

               

              But this server (the one one who acts as file server) doesn't have a NSH Console installed in it, just the RSCD, so that... according to common sense any change you could do in the "secure" file affects to the RSCD behaviour, so that there is not need to have NSH console installed make this configuration takes place, is this correct?

               

              I would also like to know if you know if there is way to set "bulk" mode instead of specific write buffer size in the secure file.

               

              My best regards.

              • 5. Re: Is there any way to optimized the performance in terms of I/O of the execution and access to files and script through NSH...?
                Bill Robinson

                from the man page you would update the buffer size on the target size - where you are writing.  so i don't think you need to do anything here to the file server agent since you are reading from there, not writing.  the optimization would be done to all of your targets.

                 

                since the buffer size is dependent on the system and network,  i don't think there would be a bulk mode.