Historic archive of defunct list bloat-devel@lists.bufferbloat.net
 help / color / mirror / Atom feed
* Re: [Cerowrt-devel] Raspberry Pi build
       [not found]   ` <502FD9EA.6070009@gmail.com>
@ 2012-08-18 18:53     ` Dave Taht
  0 siblings, 0 replies; only message in thread
From: Dave Taht @ 2012-08-18 18:53 UTC (permalink / raw)
  To: Robert Bradley; +Cc: bloat-devel, cerowrt-devel

On Sat, Aug 18, 2012 at 11:07 AM, Robert Bradley
<robert.bradley1@gmail.com> wrote:
> On 18/08/12 18:38, Dave Taht wrote:
>>
>> 2) I have no idea what wireless chip is in the pi, nor its current state.
>> (?)
>
>
> The Raspberry Pi has no built-in wireless chip, but using USB WiFi adapters
> would be possible.
>
>> 3) In order for things like fq_codel to have a hope of working well at
>> the ethernet level, "Byte Queue Limits" (BQL) is *required*. I don't
>> think any USB to ethernet adaptors have that yet, but I would welcome
>> being wrong. And even if BQL worked over USB, we might have issues
>> with the USB layer left to deal with... but BQL is easy to implement
>> on native ethernet (6-8 lines of code or so)...
>>
>> I went looking through the source trees briefly and didn't find
>> anything that jumped out at me. If you can point me to the right
>> drivers I'll take a look. (and/or ask question 3 on a rasperry pi list
>> for us?)
>>
>
> I did a bit of digging, and the Raspberry Pi uses the SMSC LAN9512 chip for
> wired Ethernet
> (http://www.smsc.com/media/Downloads_Public/Data_Sheets/9512.pdf). The
> smsc95xx driver currently lacks BQL support, so that would need to be added
> first.

22k worth of native buffering in the driver (at least at this layer)
(at 100Mbit, I usually use 3k)

#define DEFAULT_HS_BURST_CAP_SIZE       (16 * 1024 + 5 * HS_USB_PKT_SIZE)
#define DEFAULT_FS_BURST_CAP_SIZE       (6 * 1024 + 33 * FS_USB_PKT_SIZE)

And: Either this is a "yea! one abstraction to fit them all", or a
"oh, god, we need something else to track completions", as above it
uses a generic usbnet.c driver, which, sigh, does try to aggregate
too.

Interestingly it tries to live for 5ms at a time, interrupt-free. This
is worrisome given codel at the layer its at and its 5ms target.

Also wandering through the various conditions in tx_complete was "interesting"

I kind of wish I hadn't looked! But for all I know this is a simpler
use-case for addressing aggregated packets than wifi...



>
> --
> Robert Bradley
>
> _______________________________________________
> Cerowrt-devel mailing list
> Cerowrt-devel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cerowrt-devel



-- 
Dave Täht
http://www.bufferbloat.net/projects/cerowrt/wiki - "3.3.8-17 is out
with fq_codel!"

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2012-08-18 18:53 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <502F2FA3.1000004@renta.net>
     [not found] ` <CAA93jw4KTYEOMTJP7AX6DdNUTiQrT+Kw8tbDf4CuR4CVtSAPkA@mail.gmail.com>
     [not found]   ` <502FD9EA.6070009@gmail.com>
2012-08-18 18:53     ` [Cerowrt-devel] Raspberry Pi build Dave Taht

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox