7 Replies Latest reply on Apr 3, 2016 10:49 AM by PatrickRisch

    Webpublishing: extreme different response times

    PatrickRisch

      Hi

       

      I have  a setup two machine deployment for FileMaker Server 14.04 and WebDirect.

      On one machine sits FileMaker Server with the Databaseengine.

      On one machine sits the Webpublishing Engine (WebDirect, PHP; XML), etc.

       

      Both machines are virtual servers, not on the same physical Hardware.

      Windows 2012 Servers. 8GB RAM, 4 vCores CPUs.

       

      Ping from the outside to either one machine is around 30ms

      Https request take as well about 30ms

      An XML Request (always the same) takes sometimes 30ms, then 1 Minute. It is almost like a wave. (see the attachment)

      The request is very simple. a find all, one record, simple layout, one field. (this is used as a life bacon)

       

      A Ping from webdirect to the database server itself is always less than 1ms.

      So what is causing this extreme wide range of response times?

       

      Thanks for any insight into this.

      Patrick

        • 1. Re: Webpublishing: extreme different response times
          mikebeargie
          Both machines are virtual servers, not on the same physical Hardware.

          Are they at least on the same subnet? The only time I've seen network lag that bad in a 2-machine deployment was when the second server was being accessed across the WAN and not inside of a LAN.

           

          Are you testing this isolated while nothing else is going on server-side?

           

          EG all scheduled scripts, perform scripts on server, filemaker pro/go clients, etc... all disconnected?

           

          Under activity can you see the xml session connected and taking a minute? Is it possible that any other action that is happening could be causing issues. For example one solution I worked on was performing the "exit application" script step after each basic find, causing the OnLastWindowClose action to fire every time and add ~15 seconds to each PHP call. This snowballed when a few hundred users connected at once.

          • 2. Re: Webpublishing: extreme different response times
            PatrickRisch

            Hi Mike

             

            yes both machines are on the same subnet.

            The lag is all day as bad, even in the evening hours, when almost no clients are connected.

            Some PHP scripts return sometimes error 502 'bad gateway'.

             

            Patrick

            • 3. Re: Webpublishing: extreme different response times
              mikebeargie

              Have you tried running a packet sniffer on your network to see if something is blasting it with packets?

              • 4. Re: Webpublishing: extreme different response times
                CarstenLevin

                Supplementary to Mike: Round Trip Time (ping time) between the two servers. Not just once, but over time. How is it, have you measured?

                And while virtual servers are absolutely OK, we have solutions running on VM's, you need to be sure that there are no problematic bottle necks, between them as covered so far, and especially on each of them.

                 

                And how is the performance with FileMaker clients (FileMaker Pro/Advanced or FileMaker Go) connected directly to the FileMaker DB Server?

                • 5. Re: Webpublishing: extreme different response times
                  PatrickRisch

                  I just sent 2000 Pings from

                  - WAN to the database server: average 8ms

                  - WAN to the webdirect server: average 8ms

                  - webdirect server to the filemaker server: average < 1ms, maximum 8ms. (There is one hope in between (the gateway itself).

                   

                  The performance of the FileMaker Clients is good.

                  • 6. Re: Webpublishing: extreme different response times
                    CarstenLevin

                    Hi Patrick,

                     

                    Then it is obviously not at network problem.

                     

                    Here are some questions to make us able to continue:

                    • Can you make a local test on a suitable PC Server, a one machine setup for test. How is it working?
                    • Who is responsible for the Virtual machine*?

                     

                    I am pretty sure that it is something in the VM setup that is causing the problems. On the worker/WPE or the DB server.

                    Very often the VM responsible person will claim that he/she has allocated so many processors, so much RAM etc. etc. etc. and that the VM is very fast.

                    But only very few VM responsible are in fact VM experts. Being a VM Expert is not a matter of reading the manuals. It involves deep understanding of:

                    • Hardware
                    • OS
                    • VM and virtualisation
                    • The actual server/application

                    And it is more complex than that because between each point there are deep interactions which change and affect each other.

                    In most cases it is a matter of hidden bottle necks in the VM setup.

                     

                    Thus: The right way to test is to set it up on dedicated hardware/OS and test that it is working. Then you tell the VM expert "See, it is working, now you make sure that it is also working on the VM".

                     

                    If the VM environment is not able to do this, it is a problem for the VM responsible.

                     

                    Best regards

                     

                    Carsten

                     

                    *If working in a single machine setup it should work faster in a two machine setup.

                    • 7. Re: Webpublishing: extreme different response times
                      PatrickRisch

                      Thank-you both for your help.

                      I will looking into it, and let you know.

                      (and open a new thread with a related question...)