[DRBD-user] 8.x performance vs 0.7x --- was (not urgent) Request for DRBD Developers

[Tom Brown]

On Fri, 23 May 2008, David Coulson wrote:

> Gordan Bobic wrote:
>>  Personally, I'm still on 8.0.8. Later versions at the time seemed to show
>>  _VASTLY_ reduced performance on my hardware, so I excluded it from the
>>  updates. I'm not experiencing any problems, so I'm sticking with it. 
> Interesting. I've honestly not ran any performance tests on drbd82 yet - My 
> applications are 90% reads, so I don't think there is much overhead as it's 
> not touching the network.
>
> Has anyone ran real performance tests of drbd using bonnie++ or iometer and 
> compared it to raw disk/LVM/etc?

what makes an bonnie++ or iometer "real"?

I've tried the current versions of 8.0 and 8.2 and see sufficiently 
serious performance problems (on the centos 5.1 kernels) compared to the 
earlier 0.7 that I've quit running a real-time drbd mirror. system load 
averages went from aprox .5-1 to 5-10 ...

I give my machines enough memory that 80-90% of I/O is writes. IMHO trying 
the update on a real system was the ultimate "real" benchmark...  put a 
live (domU) system on it and find out what happens. In this case what 
happened was not good...

... all the above said, I did end up changing a bunch of things. I went 
from hand-built xensource kernels, which are archaic linux versions, and 
running on x86_64 to the centos i386 kernels... but it wasn't a single 
step. The one thing that was consistent was that I couldn't get .7x 
performance out of the 8.x versions... I have no idea why. I did post 
asking about it but got no responses

I plan to try the 8.0.8 version after having seen the above post.