Forum Discussion

snickerbooze_13's avatar
snickerbooze_13
Icon for Nimbostratus rankNimbostratus
Oct 26, 2011

IIS 6.0 seems slow behind BIG-IP

Short story:

 

Accessing site directly is faster than accessing through BIG-IP.

 

 

Long story:

 

I have been working through several issues with some success. The server is for reporting so some of the requests exceed 5 minutes. That was my first problem.

 

I was using the base tcp profile, but figured out that the connection was timing out when the connection was reset after 300 seconds. I created a new TCP profile with increase to 600 seconds. Problem appears to be resolved. However, this got me thinking....

 

The particular script used for testing takes just over 4 minutes when accessing directly. When i access through BIG-IP the request takes nearly 9 minutes. This is perplexing as it doesn't seem like what would be considered expected behavior.

 

So, the question is how big of a problem do i have. How many requests actually take longer than 5 minutes and out of those requests that do take longer than 5 minutes, how much of it is because it is slower because of BIG-IP?

 

With that said, it would appear that the profile setting is in need of some tuning. I notice there are other profile with the name of tcp-wan-optimized and such wich has higher threasholds in what appear to be the relevant areas.

 

My question is weather i am looking in the right place or not? I do not have access to the LB or i would check myself. I can request changes to be made, but cannot make requests one right after the other in a brute force approach.

 

 

Here is my setup:

 

browser -> BIG-IP ver 10 -> stand alone IIS server -> page_with_5_minute_query

 

 

thanks in advance, any help is appreciated.

 

 

 

7 Replies

  • Anyone have a suggestion........ It seems like a simple question that someone could at least say weather or not i am on the right track or way off.

     

     

    Hate to seem pushy, but i really need to get this figured out.

     

     

    Thanks again
  • Try tcp-lan-optimized profile and modify it to increasae the timeout as you have done before - this should address it.
  • Was it configured using the template? Are you doing SSL offloading? Have you tried shutting off Nagle's Algorithm? Have you tried using the LAN optimized TCP profile as a test?

     

     

     

    Also, read this thread, good info here:

     

     

     

    http://devcentral.f5.com/Community/...fault.aspx

     

     

     

    Good luck.

     

     

  • Sorry, i have been away for a while. Don't mean to seem like i am going to ask question and forget about it, but just got somthing that moved up a little higher on my list of priorities..... so, i still have the same issue mentioned above. I didn't change any settings yet on BIG IP because i don't feel like i have understanding of all my questions yet. I have a few more i would like to fiugure out.

     

     

    The link from steve was verry helpful and i belive there were links on the site linked to that were very useful for my under standing. That actually led me to this question......

     

     

    My windows 2003 server default tcpwindowsize in the registry exceeds what the defualt settings are with the big IP tcp profile for the

     

     

    proxy buffer low

     

    proxy buffer high

     

    send buffer

     

    recieve window

     

     

    anyway, long story short is i am trying to build some understanding of how a big ip between my browser and the web server would affect performance if the settings seem to be lower or bottle neck of sorts. Is this why there is a default big ip profile for LAN / WAN optimizaion?

     

     

    Are the values in those profiles based on theses MS default server settings or something? If I'm not mistaken i thought i saw something in my searching that there is a particular gude for setting up IIS server.

     

     

    Thanks in advance for all the help
  • Actually, this link helps answer my quesiton:

     

     

    http://devcentral.f5.com/Default.aspx?tabid=63&articleType=ArticleView&articleId=287

     

     

    anyone have somehting in particular to windows 2003 enterprise?
  • Okay, true or false:

     

     

    Scenario 1)

     

     

    windows 2003 defualt is 64240. Big IP recieve window defaut is 32768 for TCP profile.

     

     

    Question 1)

     

     

    is this telling me that the largest window that is going to be advertised is 32768 to the client even though the windows default 64240? Would that be a bottle neck of sorts?

     

     

     

     

    Scenario 2)

     

     

    proxy buffer high is default 16384. This means big ip recieve window closes at half the window size.

     

     

    Question 2)

     

     

    IS this recieve window closed same as zero window in some way?

     

     

     

    So far, all indication are that the settings on my f5 are bottle necking my server by conflicting settings. I realize this is not a knock on f5, rather my configurtaion setup not optimized. I am trying to ascetain reason behind the settings for big ip optimized profiles and how it could help performance in my case.

     

     

    I tried to use wire shark to see the window size but it is inconclusivve to me. I see window size of 64k from iis server to big ip. From big ip to server is see window in the range of 8k maybe. I am un able to determine the meaning of that at this point. If someone can help a guy out that'd be great

     

     

    thanks,
  • depends on scaling setting as well...tcp can be a beast to understand with several tweaks in the protocol assisting or countering effectiveness based on far too many attributes to cover in a forum post. I wrote an article series on the TCP protocol a while back that covers several of the settings. The one on windows and buffers is here:

     

     

    http://devcentral.f5.com/Tutorials/TechTips/tabid/63/articleType/ArticleView/articleId/287/Investigating-the-LTM-TCP-Profile-Windows-amp-Buffers.aspx

     

     

    The lan optimized profile is often (in my experience) the friendliest for wan and lan environments for standard http traffic. I'd give that one a try. Just to give you a real-word, I had an app that was "faster" without the BIG-IP by about 100%, until I put an optimized profile in place, and then the BIG-IP helped to trim the original avgl response time by about 12%.