| To: | Steve Lord <lord@xxxxxxx> |
|---|---|
| Subject: | Re: xfslogd on SMP systems |
| From: | Andi Kleen <ak@xxxxxxx> |
| Date: | Fri, 4 Apr 2003 05:28:31 +0200 |
| Cc: | "HABBINGA,ERIK (HP-Loveland,ex1)" <erik.habbinga@xxxxxx>, "'linux-xfs@xxxxxxxxxxx'" <linux-xfs@xxxxxxxxxxx> |
| In-reply-to: | <1049409123.303.6.camel@xxxxxxxxxxxxxxxxxxxx> |
| References: | <F341E03C8ED6D311805E00902761278C0C35E4BD@xxxxxxxxxxxxxxx> <1049409123.303.6.camel@xxxxxxxxxxxxxxxxxxxx> |
| Sender: | linux-xfs-bounce@xxxxxxxxxxx |
> It is being reworked. The threads handle I/O completion on a per cpu > basis, interrupts hand work off to the thread on the same cpu. Xeons > deliver all their interrupts to cpu 0 by default, so all the work > goes one place. Ask intel about this one. There are various patches to do irq balancing in software (from Intel, from Ingo Molnar, from Andreas Arcangeli). You can even do it in user space using irqbalanced. Near all vendor kernels are shipping with some version of these patches. 2.5 has it too. Only Marcelo has not yet merged them. I would rather get users to apply these patches or run irqbalanced than rework XFS. -Andi |
| <Prev in Thread] | Current Thread | [Next in Thread> |
|---|---|---|
| ||
| Previous by Date: | TAKE - pagebuf, Nathan Scott |
|---|---|
| Next by Date: | recover deleted files?, Vernon A. Fort |
| Previous by Thread: | Re: xfslogd on SMP systems, Steve Lord |
| Next by Thread: | RE: xfslogd on SMP systems, HABBINGA,ERIK (HP-Loveland,ex1) |
| Indexes: | [Date] [Thread] [Top] [All Lists] |