[Pc_Support] Re: 3Ware question -- context is everything

[Jason Boxman]

On Monday 20 February 2006 19:50, Bryan J. Smith wrote:
> Jason Boxman <jasonb at edseek.com> wrote:
> > I completely didn't follow.
>
> Yes, because you're taking the discussion out of the _original_
> context.

And I interjected a question in a relevant thread.

> The original context was:
>   3Ware on 133MBps shared PCI32 at 33 v.
>   software on bi-250MBps (500MBps effective) PCIe x1
>
> I said given a 75MBps DTR hard drive, you will get ...
>
> For an effective 75MBps write:
>   -  75MBps write using a 133MBps PCI32 at 33
>   - 150MBps write using a bi-250MBps PCIe x1
>
> For an effective 150MBps interleaved read:
>   - 150MBps read using a 133MBps PCI32 at 33 (ugh!)
>   - 150MBps write using a bi-250MBps PCIe x1

Makes sense.

> So given these facts, using a 3Ware card in a PCI32 at 33 channel makes
> _less_sense_ -- especially since you will _saturate_ your PCI bus (at
> least with modern disks and a good, interleaved set of reads).
>
> But _only_ if you can guarantee the SATA channels in the software
> RAID are using a dedicated PCIe x1 channel.  Otherwise, if you use
> SATA channels on the same PCI32 at 33, or any PATA channels (most are
> connected on the legacy PCI32 at 33), there is no dedicated PCIe x1 with
> 250MBps in each direction.

If your I/O is such that you rarely will do your DTR per drive under your I/O 
load, would it matter which solution you deploy?  It seems to be an issue of 
which bottleneck you're actually going to reach.  My drives only have DTRs of 
~ 50MB/s anyway.

-- 

Jason Boxman
http://edseek.com/ - Linux and FOSS stuff