Comments

• richard.li Members

  December 2009 edited December 2009

  The first problem found is in the ProxyProcessor.isHostTheProxy(), it will try to find the ip of the non-existing host: "sahi.example.com" and cause a large delay for the dns to timeout. I fixed it as follows:
  
  try
  {
  if (host.equals("sahi.example.com"))
  {
  return false;
  }
  
  String hostAddress = InetAddress.getByName(host).getHostAddress();
  
  return hostAddress.equals("127.0.0.1") || hostAddress.equals(InetAddress.getLocalHost().getHostAddress());
  }
  catch (Exception e)
  {
  return false;
  }
  
  But I think the bast way to do this is to define the sahi.example.com as a constant so that all codes refer to this domain can point to a single version. This will speed up page loading quite a bit.
• richard.li Members

  December 2009

  The second problem discovered is the stream filter architecture. It seems it uses a lot of piped streams and threads to implement, however the best way in doing this should be using filtered streams approached so that no unnecessary threads is required. I will try to work this out and see how much performance can be improved.
• narayan Administrators

  December 2009

  Hi Richard,
  
  Thanks for the inputs and do let us know what you find. We are ourselves looking for ways to speed up Sahi further, and your findings will be very valuable to us.
  
  ProxyProcessor.isHostTheProxy() is called only for urls which do not have /_s_/ in their urls. All calls to sahi.example.com always have _s_ in their urls so they will never reach ProxyProcessor.isHostTheProxy(). Let us know if you noticed something different.
  
  We are also looking at reducing the number of injected files at the top to reduce round trips.
  
  Do keep us posted.
  
  Thanks and regards,
  Narayan
• richard.li Members

  January 2010

  Actually I found ProxyProcessor.isHostTheProxy()'s slowness is by profiling and found that a lot of time required to execute InetAddress.getByName(host), and when I trace in, I found the host name is sahi.example.com, so there maybe some bugs in other places causing this problem.
• richard.li Members

  January 2010

  Another slowness discovered is the filter architecture used. Your design uses one thread per filter to do the copying, this cause a lot of threads intercommunication problem and slow down the filtering process. I have made a wrapper change the multi-thread model to a simple filter stream model and it speed up a lot.
  
  Another slowness is due to the SSL socket connection, I have observed some place you have not used buffered stream for I/O and cause slowness in SSL encryption, also increase the default buffer size can help a lot in this case, I have changed all bufferedstream to use 64K buffer instead of 8K buffer.
  
  The overall speedup is 5 mins vs 20-30secs
• narayan Administrators

  January 2010

  Awesome Richard! Would you be willing to share it?
• richard.li Members

  January 2010

  Sure, I have to do some code cleanup first, but how may I upload the changes? Or I may commit the change to you code base if you permitted.
• narayan Administrators

  January 2010

  You could just zip up the source and send it across to my email id. Let us review it before you check it in.
  
  Thanks again,
  Narayan
• richard.li Members

  January 2010

  Just send you an e-mail with modified sources with the reason of my changes, hope this helps.
• narayan Administrators

  January 2010

  Thanks a lot. I will keep you updated.