On 10/16/2013 03:57 PM, Prakash Surya wrote:
> On Wed, Oct 16, 2013 at 06:40:12PM +0100, Saso Kiselkov wrote:
>> On 10/16/13 6:27 PM, Prakash Surya wrote:
>>> If the completely dynamic approach isn't tractable, why split the table
>>> into a 2D array? Why not just increase the size of it, and keep it a 1D
>>> array?
>>
>> The reason why I split it was to reduce a single allocation size. The
>> point is that on bigmem machines (with 1TB+ of physical DRAM) these
>> allocations can grow to ridiculous sizes (1GB+). If the virtual address
>> space is sufficiently fragmented this can create trouble. So far, at
>> least, this is my hypothesis. If it's invalid, I will happily revert the
>> code back to a 1D table, but so far we haven't been able to get VM
>> experts to comment on this. I have seen one edge case where the kernel,
>> under memory pressure, failed to allocate a contiguous 128k buffer -
>> whether the scenario applies generally, though, I'm not certain. Again,
>> not a VM expert.
> 
> I'm no VM expert either, but speaking from Linux's perspective, I don't
> think virtual address space fragmentation is much of an issue. AFAIK,
> whether you're doing a 1M vmalloc or 1G vmalloc, VM fragmentation
> doesn't play much of an issue. The kernel will allocate non-contiguous
> pages and then present them as a contiguous region, so you just need
> enough free pages on the system to satisfy the request.
> 
> I should try and prod behlendorf about this, since he has much more
> experience on the subject than I do.
> 

It would only cost a problem on 32-bit. On 64-bit, we have more address
space than we can possibly use right now.

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
developer mailing list
[email protected]
http://lists.open-zfs.org/mailman/listinfo/developer

Reply via email to