I have read (in this forum) about people using dual quad-core servers with tons of RAM.
For Open-e, what provides a better return: CPU or RAM?
Is there an optimum level, after which there is no appreciable gain?
Printable View
I have read (in this forum) about people using dual quad-core servers with tons of RAM.
For Open-e, what provides a better return: CPU or RAM?
Is there an optimum level, after which there is no appreciable gain?
Hi
I'm the guy who had loads of memory en dual quad 5335.
After a few weeks of testing my findings about this are that open-e or the iscsi target that is used (iscsitarget)
- does not use more than one CPU
- and when using blockio memory is not used because filesystem cache is bypassed
In all my tests i've never seen the cpu usage go above 100%, ( with 8 cpu's you can go to 800%) i've seen it.
blockio is the current default and fileio is unusable because of initialization failing after 3 iscsi volumes.
My advise is to go for with and
- a fast dual single core machine
- 2~4 GB of memory
- a good raid controller ( i use a areca 1230ML 12 disk )
- good ethernet controllers ( dual intel server e1000 )
when reading large chunks of data from the open-e san i get the bonded dual e1000 full ( 200+ Mb/s)
when writing large chunks of data to the open-e san I get 100Mb/s
when reading & writing I get on 1 interface i get 100Mb in and out of it, using the full duplex
Update of my previous post
I've done some testing and i see that there is a difference in IOps when using 1 CPU instead of 8. There is a decrease in speed. my gues is 10 - 50 % drop. It depends on the size of the io size you are reading/writing.
So multiple CPU's are beter then one. How many is useful is still being tested.