Caching over SMB to a NAS is extremely slow.
3696 3 2-
- utterchaos
- Member
- 2 posts
- Joined: March 2015
- Offline
Hey so I recently upgraded everything to 10 GBE. I have a NAS that windows writes and reads to at over 350MB/s. When I cache to it though from within Houdini, (DOP Pyro VDBs), I don't seem to get anywhere near that speed. In fact it looks like it's saving files at 2MB/s. Houdini works fine when writing to local disks. What am I doing wrong?
-
- malbrecht
- Member
- 806 posts
- Joined: Oct. 2016
- Offline
Moin,
you are not doing anything wrong, but making a false assumption. 350MB/s is a completely arbitrary measurement (i.e. “academic” or “not helpful”) in that context because what you want to do with caching is “random access”, not linear reading/writing (that value is only applying to linear access with optimized block sizes). TCP/IP, the protocol for ethernet, is not really well designed for lots of small(est) packages with constant back and forth communication.
If you want performance on caches, use the fastest possible SSD or RAM drive you can get, plug it into your PCI express port and give it all the IRQs it wants. Intel just released a cool new SSD system that boosts speed on random access by factors.
Using network drives for caching is, except for huge datasets that only get accessed in an exclusively linear fashion, never going to work well.
Marc
you are not doing anything wrong, but making a false assumption. 350MB/s is a completely arbitrary measurement (i.e. “academic” or “not helpful”) in that context because what you want to do with caching is “random access”, not linear reading/writing (that value is only applying to linear access with optimized block sizes). TCP/IP, the protocol for ethernet, is not really well designed for lots of small(est) packages with constant back and forth communication.
If you want performance on caches, use the fastest possible SSD or RAM drive you can get, plug it into your PCI express port and give it all the IRQs it wants. Intel just released a cool new SSD system that boosts speed on random access by factors.
Using network drives for caching is, except for huge datasets that only get accessed in an exclusively linear fashion, never going to work well.
Marc
---
Out of here. Being called a dick after having supported Houdini users for years is over my paygrade.
I will work for money, but NOT for "you have to provide people with free products" Indie-artists.
Good bye.
https://www.marc-albrecht.de [www.marc-albrecht.de]
Out of here. Being called a dick after having supported Houdini users for years is over my paygrade.
I will work for money, but NOT for "you have to provide people with free products" Indie-artists.
Good bye.
https://www.marc-albrecht.de [www.marc-albrecht.de]
-
- kensho.murata
- Member
- 3 posts
- Joined: Oct. 2021
- Offline
-
- Jonathan de Blok
- Member
- 281 posts
- Joined: July 2013
- Offline
It depends on the workflow but another approach could be to save it locally and have some background tool push it to the NAS. That will be the fastest way to get it out of Houdini and get back into the UI.
And on a related note, when saving the hip file on the network gets slow check the HOUDINI_BUFFEREDSAVE env var options.
And on a related note, when saving the hip file on the network gets slow check the HOUDINI_BUFFEREDSAVE env var options.
More code, less clicks.
-
- Quick Links