Will send those too. The idea is to let kernel consumers enjoy the improvements
to latency that blue flame gives. And yes, SDP is motivating us but I am going
to push to IPoIB too.
I want to take the opportunity that you raised the issue to hear others opinion
about changing the bitmap
Eli Cohen wrote:
I was going to send [...] upstream
Also you had a fix to the port speed and something related to SL which I
didn't understand, please send for review
Or.
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to
Eli Cohen wrote:
The idea is to let kernel consumers enjoy the improvements to latency that blue
flame gives. And yes, SDP is motivating us but I am going to push to IPoIB too.
From my recollection of numbers, for user space apps, using inline
accounts for about 1us improvement in the
On Mon, Nov 08, 2010 at 10:27:15AM +0200, Or Gerlitz wrote:
Is this correct that today all the kernels QPs use the same UAR, any
problem with that?
Yes, today all kernel consumers use the same UAR. And if you want to
use blue flame then you need to use different UARs.
--
To unsubscribe from
Or Gerlitz wrote:
Eli Cohen wrote:
From my recollection of numbers, for user space apps, using inline
accounts for about 1us improvement in the latency, if this is indeed the
case, I'm sure if there's great value here for kernel consumers, do you
have any numbers to support this patch?
I
On Mon, Nov 08, 2010 at 10:24:26AM +0200, Or Gerlitz wrote:
Eli Cohen wrote:
I was going to send [...] upstream
Also you had a fix to the port speed and something related to SL
which I didn't understand, please send for review
There have been a few minor fixes - I'll review and send.
--
To
On Mon, Nov 08, 2010 at 10:45:42AM +0200, Or Gerlitz wrote:
Or Gerlitz wrote:
Eli Cohen wrote:
From my recollection of numbers, for user space apps, using inline
accounts for about 1us improvement in the latency, if this is indeed the
case, I'm sure if there's great value here for
Eli Cohen wrote:
It indeed improves SDP's latency - I don't have exact numbers.
the SDP number is very interesting (Amir, do you have it?) but irrelevant for
upstream, any IPoIB numbers?
Or.
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to
I am working on it, will notify when have numbers.
On 11/08/2010 10:51 AM, Or Gerlitz wrote:
Eli Cohen wrote:
It indeed improves SDP's latency - I don't have exact numbers.
the SDP number is very interesting (Amir, do you have it?) but irrelevant for
upstream, any IPoIB numbers?
Or.
--
To
Hefty, Sean wrote:
CM mads aren't reliable, however they are retried. If a CM REQ does not
receive a response after so many retries (usually 15), the REQ fails (status is
timeout). The mad layer reports the timeout to the cm module. With snooping
in place, a user will be notified that a
Ethtool documentation states that when one of the parameters, rx_coalesce_usecs
or rx_max_coalesced_frames is set to zero while the other has a none zero
value, the none zero parameter should still be operative. For example, if
rx_max_coalesced_frames is set to zero while rx_coalesce_usecs is 0,
On Mon, Nov 08, 2010 at 10:51:30AM +0200, Or Gerlitz wrote:
Eli Cohen wrote:
It indeed improves SDP's latency - I don't have exact numbers.
the SDP number is very interesting (Amir, do you have it?) but irrelevant for
upstream, any IPoIB numbers?
For IPoIB it gives ~1 usec for
Eli Cohen wrote:
For IPoIB it gives ~1 usec for improvement in latency.
yep, this is what I expected, so over your testbed from what value to what
value? also it
would be important to note the change in the cpu utilization (e.g few vmstat
1 output
lines before/after the change, while running
so the usage of mad snooping would be for cache invalidations, I wonder
if registering on GID/MGID IN/OUT traps be sufficient for the same purpose?
That requires registration with the SA. The intent is to avoid using a
centralized service when possible. Otherwise, we end up with all nodes
Roland,
this patch has bad affect on mlx4_en. I sent another patch which does
the same at the IPoIB layer. Please consider removing it.
- thanks.
On Mon, Oct 25, 2010 at 04:38:59PM +0200, Eli Cohen wrote:
The modify CQ command disables the effect of both the count and period
paramters if any
Hefty, Sean wrote:
That requires registration with the SA. The intent is to avoid using a
centralized service when possible.
yep, makes sense, look like this design finally went the decentralized way...
cool
Or.
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the
this patch has bad affect on mlx4_en. I sent another patch which does
the same at the IPoIB layer. Please consider removing it.
I didn't merge this yet, did I?
- R.
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
I wonder if anyone has ever looked at this? It looks like there is
something wrong here.
I was just looking at some results from a stock 2.6.35 kernel doing
TCP over IPoIB (2k MTU) and CUBIC looks like it is going nutz, you can
see in a bandwidth plot a fairly characteristic saw tooth pattern,
On Tue, Nov 9, 2010 at 1:04 AM, Roland Dreier rdre...@cisco.com wrote:
I didn't merge this yet, did I?
No, you did not. I was just asking that you won't.
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo
19 matches
Mail list logo