[openib-general] IPoIB IETF WG presentation updated again

[Dror Goldenberg]

> -----Original Message-----
> From: Roland Dreier [mailto:roland at topspin.com]
> Sent: Tuesday, August 03, 2004 5:24 PM
> 
> 
>     Hal> Cloning an AH is probably faster than creating a new one from
>     Hal> scratch. (We would need an additional verb for this). How
>     Hal> much does this cost ? Is this optimization worth it ?
> 
> Actually cloning an AH is probably slower on Tavor than creating one
> from scratch, at least if the AH is in Tavor memory.  To clone an AH
> one has to do a bunch of reads across PCI and then write the data back
> across PCI, while creating a new one is just a matter of taking the
> address attributes (which are almost certainly already in CPU cache)
> and writing them across PCI.  (Reads across PCI are much more
> expensive than writes, which can be posted and even write-combined)
> 
If you want to do a fair comparison, then you need to compare ModifyAH
with DestroyAH+CreateAH. Anyway, since AVs are read only from the HCA 
perspective, if you want to do a good implementation of AV operations, then 
you can completely avoid PIO reads. Just maintain a shadow data structure
in the process address space that will hold the kind of information that
you intended to read from the device memory. While this solution requires
more memory, this memory doesn't need to be pinned, so it's relatively 
a cheap resource and it reduces PIO reads completely.

Dror
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20040805/a10c49cc/attachment.html>