5 Replies Latest reply on Mar 9, 2014 6:03 AM by disabled_ScottKoontz

    Server "power" & network speed/hops, an odd comparison

      A Mac Pro 15 hops away vs a Mac Mini zero hops away. Not your typical comparison, but it does help point out FileMaker's strengths and weaknesses. 70 million records, 20 GB. A notable difference is that the Pro is running FMS 13, while the Mini has FMS 12.

       

      The reason we purchased the MP was to replace a Mac Mini that was temporarily serving a huge solution to several/many remote users. The old server is a 2.5 GHz Mac Mini with 8 GB of RAM, and it does surprisingly well. It was originally set up to test the feasibility of hosting such a large file, and did the job so well that we served the file with this anemic machine for a year. The Pro is a 3.7 GHz Quad Core, 32 RAM and 512 SSD, sitting on a 100-down 20-up cable connection. Kudos to FileMaker for allowing such a monster DB to be hosted on a cheap machine with few problems (edit: I'm speaking of the Mac Mini, and the clients are happy with this server, but not me.)

       

      A general summary:

      1. Finds and sorts on one large table (no related fields involved) are typically a little faster from a remote Mac Pro when compared to a local Mini. I'm assuming disk access and RAM are the reason, but both machines perform well on a 70 million record table, but surprised that a fast server can overcome much slower network access. Initial sorts can be up to 2x faster on the remote Pro, but there seem to be some periodic burps on the Mini that cause a larger variance.
      2. Initial searches that include a field or two from a remote table (one-key relationship) are slightly faster on the remote Mac Pro vs local Mac Mini. Again there must be disk activity that trumps data transfer needs. Server seems to be handling much of a standard search.
      3. Wildcard searches on the remote Pro were much faster. MUCH faster. Even some refinds on the Mini were slower than the first search on the Pro. Obviously FMS is handling much of the processing work.
      4. Sorting on a field in the local table was, again, faster on the remote Pro.
      5. Sorting on a related table seemed to be one of the most network dependent things one can do. Local Mini trumps remote Pro in all cases.
      6. As I worked on both files, the Pro didn't waver and seemed to get faster, as we can expect from caching. The Mini seemed to get slower, probably disk access from less RAM and what must be cache swapping.

       

      One thing the Mac Pro will allow us to do is move more fields into the large primary table and rely less on related tables. We have the luxury of working with (mostly) static data, so a trend towards a larger single table is better than proper relational programming. When a sort on a local table takes a few seconds but the same sort one table removed takes minutes, it's worth breaking some rules and creating an even larger monster file.

       

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

      Bought a new Mac Pro. There's little bad that can be said of a machine that can boast some of the best components with the best connections between those components. Individually, SSD, lots of RAM, and fast processors are a bonus to an existing setup, but when you put them together there's a surprising made-for-each-other combination bonus that I expected, but not to this degree.

       

      Restarts aren't fast, they're wicked fast. Same with file compress or decompress, software install, etc. And yes, these are things that aren't done all the time or affect your clients, but it's still a joy to sit behind one of these works-of-art and perform just about any function.

       

      Backups on the Mini took a long LONG time to perform, and sometimes resulted in the Admin Console becoming unresponsive. Compressing the file, once each month, took over 5 hours. On a fresh restart, you could hardly tell the server was a toy by today's standards. BTW, the new Mac Pro compressed the same file in 40 minutes -- that's 8 x faster.

       

      The biggest surprise to me comes from a comparison between times when running large reports from the single large table (no finds or sorts to fields in other tables.) Server caching and client caching are understood and noted in some timed tests, but client reporting to the Mac Pro [15 hops and 50ms away] is almost as fast -- and sometimes faster -- when compared to reporting on the Mac Mini that's hosted in the office where I am sitting. Once you introduce a sort that requires a simple relationship, the local machine trumps the remote server.