Blogbench RAID benchmarks
dillon at apollo.backplane.com
Mon Jul 11 16:37:41 PDT 2011
:> I'm not sure where the concurrency is breaking down on the RAID subsystem,
:> it could be due to that particular driver still being under the big
:> giant lock (the AHCI and SILI drivers don't use the bgl, but arcmsr
:> driver probably still does).
:arcmsr uses lockmgr and is not under the mplock at all. Which is not
:to say it isn't still serializing internally somewhere, but it should
:not be any different than FreeBSD in this regard.
Hmm. There are two major locking points for a (CAM based) disk driver.
The first is the lock passed in cam_sim_alloc(). The second is whether
INTR_MPSAFE is passed to bus_setup_intr().
It looks like arcmsr is passing its own lock to cam_sim_alloc(), but
is *not* passing the INTR_MPSAFE flag to bus_setup_intr().
I haven't reviewed the interrupt handling code to determine how safe
the functions it calls internally are. arcmsr does get its own lock
in arcmsr_intr_handler(). It could very well be safe for us to set
the INTR_MPSAFE flag there but someone will have to life-test it.
In anycase, I'm not sure it's even the bottleneck for the earlier tests
since blockbench didn't run long enough (with default options) to use
up 32G of ram.
<dillon at backplane.com>
More information about the Kernel