[openib-general] How about ib_send_page() ?

[Jeff Carr]

Grant Grundler wrote:

> 4K -> 1.8 GB/s
> 16k -> 3.3 GB/s
> 64k -> 3.8 GB/s
> 
> This seems reasonable.
> IIRC the ZX1 chipset has 6GB/s backplane but one CPU can only drive ~4GB/s.

I have a E7501. Thanks for running this test. I'd not looked so closely 
at this before or been up to the wall against it where it matters.

1K   ->  .5 GB/s
4K   -> 1.2 GB/s
16K  -> 1.7 GB/s
32K  -> 1.8 GB/s
64K  -> 1.9 GB/s
128K -> 1.9 GB/s
256K -> 1.8 GB/s
512K -> 1.7 GB/s
1M   -> 1.2 GB/s
2M   ->  .7 GB/s

> I don't see why not.
 > It ovbiously helps on the IA64 box.
 > We want to measure the copy speed, not the syscall speed, right? :^)

Maybe someday I'll define PAGE_SHIFT to 14 and see if it boots. ia64 
does something with KERNEL_STACK_SIZE_ORDER too. Anyway, this is OT to IB.


> BTW, can you remind me again why this was important to rdma_lat test?

I didn't bring it up for that purpose; I was just speaking in general IB 
terms.

> It was just to prove the VM/memcopy wasn't the bottleneck, right?

Ya, I was just investigating these details after I noticed that raw 
memory copy numbers were not *too* far away from how far IB is supposed 
to be.

Jeff