Re: [TSO]-Refer to "State of TCP segmentation offload to NIC in linux"

Re: [TSO]-Refer to "State of TCP segmentation offload to NIC in linux"

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 762 bytes --]

On Sun, 2009-05-24 at 18:38 +0700, Nguyen Tuan Huy (G11) wrote:
> Dear Mr Ben Hutchings, 
> 
>  
> 
> I have read your answer about question "State of TCP segmentation
> offload to NIC in linux". 
> 
> The link is
> http://kerneltrap.org/mailarchive/linux-netdev/2008/5/7/1757954/thread

Please direct your queries to the netdev list, not to me personally.

> I am now writing network device driver that support TSO function. But
> when I send one packet with 64K, then TCP has divided 64K, not passed
> the whole 64K to NIC driver.
[...]

I think you won't see 64K packets for a TCP connection that was just
opened, due to the slow start algorithm.

Ben.

-- 
Ben Hutchings
Teamwork is essential - it allows you to blame someone else.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

RE: [TSO]-Refer to "State of TCP segmentation offload to NIC in linux"

[Nguyen Tuan Huy (G11)]

Thanks Ben.

>I think you won't see 64K packets for a TCP connection that was just
>opened, due to the slow start algorithm.

I have dumped the packets that passed to the hardware. I found that
these packets are always MTU size. 

I think that it must have one flag indicate that when it is turn on,
packet will not framed by TCP/IP layer, but hardware. 
At first I think that is NETIF_F_TSO, but it is not correct.

My kernel version is 2.6.11.12

Best regards,
Huy

-----Original Message-----
From: Ben Hutchings [mailto:ben@decadent.org.uk] 
Sent: Sunday, May 24, 2009 8:16 PM
To: Nguyen Tuan Huy (G11)
Cc: netdev@vger.kernel.org
Subject: Re: [TSO]-Refer to "State of TCP segmentation offload to NIC in
linux"

On Sun, 2009-05-24 at 18:38 +0700, Nguyen Tuan Huy (G11) wrote:
> Dear Mr Ben Hutchings, 
> 
>  
> 
> I have read your answer about question "State of TCP segmentation
> offload to NIC in linux". 
> 
> The link is
> http://kerneltrap.org/mailarchive/linux-netdev/2008/5/7/1757954/thread

Please direct your queries to the netdev list, not to me personally.

> I am now writing network device driver that support TSO function. But
> when I send one packet with 64K, then TCP has divided 64K, not passed
> the whole 64K to NIC driver.
[...]

I think you won't see 64K packets for a TCP connection that was just
opened, due to the slow start algorithm.

Ben.

-- 
Ben Hutchings
Teamwork is essential - it allows you to blame someone else.