I asked here a week or so ago, and referenced an older bug, but haven't seen any comments. Googling, I see others asking about the same. Datum: *every* one I've found is like this: they're all HP DL580's or DL380's, or such.
This is being used for heavy-duty scientific computing. Does anyone know if it *will* it seriously affect throughput speed if I turn off irqbalance?
It's running 6.4, btw.
mark
----- Original Message ----- | I asked here a week or so ago, and referenced an older bug, but | haven't | seen any comments. Googling, I see others asking about the same. | Datum: | *every* one I've found is like this: they're all HP DL580's or | DL380's, or | such. | | This is being used for heavy-duty scientific computing. Does anyone | know | if it *will* it seriously affect throughput speed if I turn off | irqbalance? | | It's running 6.4, btw. | | mark
That really depends on the type of computing. If it's memory or CPU intensive computing it should have little to no affect. However, if you are doing heavy I/O intensive operations, more specifically network intensive, then yes, disabling irqbalance will have a fairly significant affect on performance for those operations.
On Thu, Aug 15, 2013 at 7:32 AM, m.roth@5-cent.us wrote:
I asked here a week or so ago, and referenced an older bug, but haven't seen any comments. Googling, I see others asking about the same. Datum: *every* one I've found is like this: they're all HP DL580's or DL380's, or such.
This is being used for heavy-duty scientific computing. Does anyone know if it *will* it seriously affect throughput speed if I turn off irqbalance?
It's running 6.4, btw.
mark
You referred to http://bugs.centos.org/view.php?id=6319 in your earlier post. If you believe you are hit by the bug reported there, I strongly urge you to test the centosplus kernel that has a patch (see note 17700 of the bug report).
Regarding the current status, upstream BZ ( https://bugzilla.redhat.com/show_bug.cgi?id=887006 ) is open to the public. It is "on QA" at the moment. I suspect the fix will be in EL6.5 and possibly in a 6.4 kernel update.
Akemi
Hi, Akemi,
Akemi Yagi wrote:
On Thu, Aug 15, 2013 at 7:32 AM, m.roth@5-cent.us wrote:
I asked here a week or so ago, and referenced an older bug, but haven't seen any comments. Googling, I see others asking about the same. Datum: *every* one I've found is like this: they're all HP DL580's or DL380's, or such.
This is being used for heavy-duty scientific computing. Does anyone know if it *will* it seriously affect throughput speed if I turn off irqbalance?
It's running 6.4, btw.
You referred to http://bugs.centos.org/view.php?id=6319 in your earlier post. If you believe you are hit by the bug reported there, I strongly urge you to test the centosplus kernel that has a patch (see note 17700 of the bug report).
I'm checking with the system co-owner (the other co-owner is another Institute...), and that may take a while, because they've set up samba on it, so there's a number of folks hitting this box.
Regarding the current status, upstream BZ ( https://bugzilla.redhat.com/show_bug.cgi?id=887006 ) is open to the public. It is "on QA" at the moment. I suspect the fix will be in EL6.5 and possibly in a 6.4 kernel update.
I see in the last message that I could set intremap=off on the grub kernel command line. It then mentions "If intr-remap is turned off, you'll also have to add the kvm parameter to the boot line, or in an /etc/modprobe.d file, allow_unsafe_assigned_interrupts . Otherwise, kvm device assignment will fail.". Now, the only kvm I know is the virtual machine package; as there are no VMs running on this system, is this relevant to me?
mark
-
Akemi Yagi -
James A. Peltier -
m.roth＠5-cent.us