I just noticed this change.  Two comments:

 - The overhead is not as trivial as bug #706058 seems to imply.  If you
look at eg <linux/cpumask.h> you see that increasing NR_CPUS above
BITS_PER_LONG (ie going over 64) changes the cpumask code,
arch/x86/include/asm/spinlock.h uses different code for NR_CPUS >= 256,
etc.  I think it would be worth it for someone to run some benchmarks
and make sure the overhead is tolerable for the common case.

- I don't think the description is accurate -- don't systems with more
CPUs still boot, just without being able to use all the CPUs?  I know
having more kernel variants is painful but it might be a better idea to
have a "huge SMP" variant for the tiny fraction of systems Ubuntu is
used on that have more than 64 CPUs.  (What are the platforms that this
is targeted at?  are there really workstations with > 64 CPUs?  I can
only think of server-class HW: 8-socket Nehalem EX (8 sockets, 8 cores,
2 SMT) or 4+ socket Westmere EX (10 cores/socket 2x SMT).

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/737124

Title:
  Support workstations with greater than 64 cores

-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to