[dpdk-dev] [PATCH v2 00/41] Memory Hotplug for DPDK

Shreyansh Jain shreyansh.jain at nxp.com
Thu Mar 29 12:57:31 CEST 2018


Hello Anatoly,

> -----Original Message-----
> From: Burakov, Anatoly [mailto:anatoly.burakov at intel.com]
> Sent: Tuesday, March 20, 2018 3:38 PM
> To: Shreyansh Jain <shreyansh.jain at nxp.com>
> Cc: dev at dpdk.org; Thomas Monjalon <thomas at monjalon.net>
> Subject: Re: [dpdk-dev] [PATCH v2 00/41] Memory Hotplug for DPDK
> 
> On 19-Mar-18 8:58 AM, Shreyansh Jain wrote:
> > Hi Anatoly,
> >

[...]

> >
> > [...]
> >
> > -
> > Shreyansh
> >
> 
> Hi Shreyansh,
> 
> It will be fixed before merge, yes. I would expect this code to arrive
> in Github in the next few days.

I have been integrating with this patchset for considerable time now - attempting to get it working on dpaa2 and dpaa.

Thanks to your quick responses on IRC, at least the VA part of DPAA2 is functional now. 

Unfortunately, I am observing a degradation in performance with the patches.

Compared to                   |  % Impact
master (20526313ba4)          |
----------------------------------------------
DPAA2 IOVA=VA (hotplug patch) |  ~ -4.0%
DPAA2 IOVA=VA --legacy-mem    |  ~ -4.5%
DPAA2 (Physical)              |  Pending
DPAA  (Physical)              |  Pending

Theoretically, I agree with the discussion on a similar observation in [1] - but, at least as of writing this, I have no idea why the drop is occurring.

[1] http://dpdk.org/ml/archives/dev/2018-March/093685.html

As a side note - the base for hotplug memory patches and master has some differences (since c06ddf9698e0) - though, I don't see any patch in there which might impact performance. But, it might be good idea to rebase - just to remove that doubt.

I noticed that you have already fixed the issues related to IOVA==PA mapping in your recent code. I haven't had the chance to completely integrate my code against those. I will look through the patches in coming days.

Regards,
Shreyansh


More information about the dev mailing list