Question

2016 64-bit version slower than 2015 32-bit

  • 26 April 2016
  • 7 replies
  • 0 views

Badge

I recently installed 2016 64-bit FME. I had previously been running 2015 32-bit. I was a bit leery about going to 64-bit as I primarily write out to Esri formats, but after doing quite a bit of research I verified the readers/writers I use will work with 64-bit. That said, 2016 64-bit did work. I tested it on a workflow that pulls data from 16 feature types, writes out 6 feature classes into our SDE environment (72,163 total features), and contains 128 transformers. It took 10 min 19.5 sec, and the peak process memory usage was 1,319,888 kB. The problem is that I also ran it in FME 2015 32-bit and it took 8 min 37.3 sec and the peak process memory usage was 1,095,360. As another test, I installed FME 2016 32-bit. This process took 9 minutes 34.6 seconds and the peak process memory usage was 1,088,980 kB. Does anyone have any thoughts as to why the 64-bit version took quite a bit longer to run? Also, why might 2016 be slower than 2015? This really seems counter intuitive. Thank you for any suggestions you might have!


7 replies

Badge +2

I found the opposite when testing a 750,000 polygons through a simple workspace containing AreaOnAreaOverlayer.

  • FME 2015/2016 32 bit did not finish
  • FME 2015 64 bit 17 minutes
  • FME 2016 64 bit 9 minutes
16GB RAM Windows 7 PCCan you rule out the environment it is running in? e.g. network load, DB busy etc. In my case it was local FFS to FFS just for testing purposes. I'd log with Safe as they always say later versions shouldn't be slower.
Userlevel 4

Hi

Just a short remark: 32-bits versus 64-bits versions of any software isn't so much about relative performance, but about the maximim amount of adressable memory. It is mainly with workspaces that consume a lot (>2GB) of memory where you'll get a noticable improvement using the 64-bit version.

David

Badge +2

Hi

Just a short remark: 32-bits versus 64-bits versions of any software isn't so much about relative performance, but about the maximim amount of adressable memory. It is mainly with workspaces that consume a lot (>2GB) of memory where you'll get a noticable improvement using the 64-bit version.

David

Great point! In my case it was about memory.

Userlevel 3
Badge +13

So from the original post we have:

FME Version32/64Time201532 8:37201632 9:3420166410:19

I'm not at all surprized that the 64 bit took longer - all internal structures are effectively twice the size in a 64 bit situation, so there is more moving around. As others have said, you only win in 64 bit when you have large amounts of data piling up -- then it cleans up.

But I don't like the difference between the two 32 bit times. If you're able to send the workspace in to support@safe.com we'd like to take a look. Ideally we'd ask you to "record" (via Recorders) the input data so we could play it back here without your SDE.

Thanks for posting.

Userlevel 4
Badge +25

Hi

Just a short remark: 32-bits versus 64-bits versions of any software isn't so much about relative performance, but about the maximim amount of adressable memory. It is mainly with workspaces that consume a lot (>2GB) of memory where you'll get a noticable improvement using the 64-bit version.

David

Yes. If you use 32-bit and see this message in the log: "ResourceManager: Optimizing Memory Usage. Please wait..." then you know it's time to upgrade to 64-bit. If you don't see that message, 64-bit isn't likely to make a huge performance difference.

We have a chapter in our advanced training about performance, including how to interpret a log file and when to use 64-bit. I also took the key points from that chapter and put them into a recent blog post, although that didn't talk about 64-bit as much.

Badge

Hi

Just a short remark: 32-bits versus 64-bits versions of any software isn't so much about relative performance, but about the maximim amount of adressable memory. It is mainly with workspaces that consume a lot (>2GB) of memory where you'll get a noticable improvement using the 64-bit version.

David

Thank you @david_r for helping me understand this better. @Mark2AtSafe, thank you for the article. I

Badge

So from the original post we have:

FME Version32/64Time201532 8:37201632 9:3420166410:19

I'm not at all surprized that the 64 bit took longer - all internal structures are effectively twice the size in a 64 bit situation, so there is more moving around. As others have said, you only win in 64 bit when you have large amounts of data piling up -- then it cleans up.

But I don't like the difference between the two 32 bit times. If you're able to send the workspace in to support@safe.com we'd like to take a look. Ideally we'd ask you to "record" (via Recorders) the input data so we could play it back here without your SDE.

Thanks for posting.

Thank you @daleatsafe. I will work on assembling the workspace and data.

Reply