diff mbox series

[net-next,2/5] ionic: smaller coalesce default

Message ID 20200831233558.71417-3-snelson@pensando.io
State New
Headers show
Series ionic: struct cleanups | expand

Commit Message

Shannon Nelson Aug. 31, 2020, 11:35 p.m. UTC
We've found that a smaller default value for interrupt coalescing
works better for latency without hurting general operations.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
---
 drivers/net/ethernet/pensando/ionic/ionic_dev.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Jakub Kicinski Aug. 31, 2020, 11:50 p.m. UTC | #1
On Mon, 31 Aug 2020 16:35:55 -0700 Shannon Nelson wrote:
> diff --git a/drivers/net/ethernet/pensando/ionic/ionic_dev.h b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
> index 9e2ac2b8a082..2b2eb5f2a0e5 100644
> --- a/drivers/net/ethernet/pensando/ionic/ionic_dev.h
> +++ b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
> @@ -16,7 +16,7 @@
>  #define IONIC_DEF_TXRX_DESC		4096
>  #define IONIC_LIFS_MAX			1024
>  #define IONIC_WATCHDOG_SECS		5
> -#define IONIC_ITR_COAL_USEC_DEFAULT	64
> +#define IONIC_ITR_COAL_USEC_DEFAULT	8

8 us interrupt coalescing does not hurt general operations?! No way.

It's your customers who'll get hurt here, so your call, but I seriously
doubt this. Unless the unit is not usec?
Shannon Nelson Sept. 1, 2020, 4:20 a.m. UTC | #2
On 8/31/20 5:16 PM, David Miller wrote:
> From: Jakub Kicinski <kuba@kernel.org>
> Date: Mon, 31 Aug 2020 16:50:54 -0700
>
>> On Mon, 31 Aug 2020 16:35:55 -0700 Shannon Nelson wrote:
>>> diff --git a/drivers/net/ethernet/pensando/ionic/ionic_dev.h b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
>>> index 9e2ac2b8a082..2b2eb5f2a0e5 100644
>>> --- a/drivers/net/ethernet/pensando/ionic/ionic_dev.h
>>> +++ b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
>>> @@ -16,7 +16,7 @@
>>>   #define IONIC_DEF_TXRX_DESC		4096
>>>   #define IONIC_LIFS_MAX			1024
>>>   #define IONIC_WATCHDOG_SECS		5
>>> -#define IONIC_ITR_COAL_USEC_DEFAULT	64
>>> +#define IONIC_ITR_COAL_USEC_DEFAULT	8
>> 8 us interrupt coalescing does not hurt general operations?! No way.
>>
>> It's your customers who'll get hurt here, so your call, but I seriously
>> doubt this. Unless the unit is not usec?
> Agreed, 8usec is really really low.  You won't get much coalescing during
> bulk transfers with a value like that, eliminating the gain from coalescing
> in the first place.

Thanks.  I'll drop this patch and come back to this issue when we get a 
chance to add adaptive coalescing.

sln
diff mbox series

Patch

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_dev.h b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
index 9e2ac2b8a082..2b2eb5f2a0e5 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_dev.h
+++ b/drivers/net/ethernet/pensando/ionic/ionic_dev.h
@@ -16,7 +16,7 @@ 
 #define IONIC_DEF_TXRX_DESC		4096
 #define IONIC_LIFS_MAX			1024
 #define IONIC_WATCHDOG_SECS		5
-#define IONIC_ITR_COAL_USEC_DEFAULT	64
+#define IONIC_ITR_COAL_USEC_DEFAULT	8
 
 #define IONIC_DEV_CMD_REG_VERSION	1
 #define IONIC_DEV_INFO_REG_COUNT	32