Net:HTTP performance downloading large files

Hi folks,
I'm working on my first ruby/rails project and have run into my first
major problem.
My application uses a web api for a scientific database and I am trying
to download a very large file(150mb) using Net:HTTP.post_form. This
operation is turning out to be very slow.

Net:HTTP is at least 10x slower than using my browser to download this
file on my local network. It also pegs one of my processors while doing
it.

It seems to have something to do with a "green thread" issue explained
here :

Are there any alternatives to using Net:HTTP to download files off the
web with ruby?

---------- Code in question -----------
  def Metacat.read(docid)
    #load uri set in environment.rb
    uri = URI.parse(Path_to_metacat)
    #uri.query = "action=read&qformat=xml&docid=#{docid}"
    response = Net::HTTP.post_form(uri, {
      'action' => 'read',
      'qformat' => 'xml',
      'docid' => docid
    })
    #this line will raise an exception if post failed
    response.value
    if(response.content_type == "text/xml")
      doc = REXML::Document.new(response.body)
      #check to see if Metacat is sending an error message or EML
      if(doc.root.name == 'error')
        nil
      else
        Eml.new(response.body)
      end
    elsif(response.content_type == "text/plain")
      DataTable.new(docid, response.body)
    end
  end

···

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

File I'm trying to download :
http://data.piscoweb.org/catalog/metacat?action=read&qformat=xml&docid=HMS001_020ADCP019R00_20060612.40.1

--
Posted via http://www.ruby-forum.com/.

How about just calling out to curl?

···

--
Craig Beck

AIM: kreiggers

It probably has more to do with the buffer size used in Net::HTTP.
Check out this thread:

  http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/202223

Hope that helps!

···

On Thu, Nov 02, 2006 at 09:43:30AM +0900, Chad Burt wrote:

Hi folks,
I'm working on my first ruby/rails project and have run into my first
major problem.
My application uses a web api for a scientific database and I am trying
to download a very large file(150mb) using Net:HTTP.post_form. This
operation is turning out to be very slow.

Net:HTTP is at least 10x slower than using my browser to download this
file on my local network. It also pegs one of my processors while doing
it.

It seems to have something to do with a "green thread" issue explained
here :
Headius: 06/01/06

--
Aaron Patterson
http://tenderlovemaking.com/

Chad;

This might be a stupid question, but it's always worth asking just in
case =). Are you sure that the performance problem in the code above
is in fetching the document. A 150M xml file can take a long time to
parse into a REXML document.

Actually, in a quick read of this:

  if(response.content_type == "text/xml")
      doc = REXML::Document.new(response.body)
      #check to see if Metacat is sending an error message or EML
      if(doc.root.name == 'error')
        nil
      else
        Eml.new(response.body)
      end
    elsif(response.content_type == "text/plain")
      DataTable.new(docid, response.body)
    end
  end

It doesn't look like your using 'doc' to do anything except check the
root node. Meanwhile REXML has to parse the entire document--tree
parser. You might want to give one of the streaming parsers a shot.

···

--
Lou.

  http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/202223

Thanks. Found this file:

and changed

    def rbuf_fill
      timeout(@read_timeout) {
        #changed by cburt
        @rbuf << @io.sysread(1024)
      }
    end

to

    def rbuf_fill
      timeout(@read_timeout) {
        #changed by cburt
        @rbuf << @io.sysread(16384)
      }
    end

Now downloading a 150MB file takes 25 seconds compared to 21 seconds for
straight curl and 40-sum seconds for curl using a popen.

The problem now is that I have a web API client that I was going to
package into a ruby-gem that would have been easy to install. Now I have
to tell people to start hacking the standard library if they want to use
it. Uhhg!

···

--
Posted via http://www.ruby-forum.com/\.

sorry meant to say I modified this file:
/usr/local/lib/ruby/1.8/net/protocol.rb

···

--
Posted via http://www.ruby-forum.com/.

With Ruby's open classes, you shouldn't have to. At the top of your
file/library/program, just open the class you'd like to modify, in
this case the module Net and the class BufferedIO, and do what you
want.

Something like:
module Net
  class BufferedIO
    def rbuf_fill
      timeout(@read_timeout) {
        #changed by cburt to a much larger buffer for speed
        @rbuf << @io.sysread(16384)
      }
    end
  end
end

Note: Modifying the standard library is usually considered bad form,
but if you know what you're doing and are explicit about it it's
usually ok.

HTH,
Keith

···

On 11/14/06, Chad Burt <chad@underbluewaters.net> wrote:

Thanks. Found this file: /usr/local/lib/ruby/1.8/net/protocol.rb
and changed
...
The problem now is that I have a web API client that I was going to
package into a ruby-gem that would have been easy to install. Now I have
to tell people to start hacking the standard library if they want to use
it. Uhhg!

That worked great Keith, thanks.

···

--
Posted via http://www.ruby-forum.com/.

Works great with me too!

···

--
Posted via http://www.ruby-forum.com/.