[dpdk-dev] [dpdk-users] Issue in draining packets when head is null

[dpdk-dev] [dpdk-users] Issue in draining packets when head is null

[Suraj R Gupta]

Hi all,
This is regarding an issue in the draining packet using rte_reorder_drain.
Suppose we have a window size of 16 and we get packets with sequence no
like 0,1,2,4,5,6. On performing drain operation, I retrieve packets like
0,1,2.
Since now order_buf->entries[order_buf->head] i.e. order_buf->entries[3] is
null i won't be able to any more like 4,5,6. I would be able to do it only
when I get at a packet with a sequence number above window size. This may
be highly undesirable if we have large window sizes say 1024, 2048, etc. we
wouldn't be able to drain any more packets until a packet above window size
is received. Also since the struct rte_rorder_buffer are defined in the
source .c file, we are not able to move head of buffer from applications. *Is
there any specific reasons why it is implemented like this in the library.
Are there any ways to overcome this.*
-- 
Thanks and Regards
Suraj R Gupta

Re: [dpdk-dev] [dpdk-users] Issue in draining packets when head is null

[Suraj R Gupta]

Folks,
Any Ideas on this.

On Mon, Jan 13, 2020 at 1:04 PM Suraj R Gupta <surajrgupta@iith.ac.in>
wrote:

> Hi all,
> This is regarding an issue in the draining packet using rte_reorder_drain.
> Suppose we have a window size of 16 and we get packets with sequence no
> like 0,1,2,4,5,6. On performing drain operation, I retrieve packets like
> 0,1,2.
> Since now order_buf->entries[order_buf->head] i.e. order_buf->entries[3]
> is null i won't be able to any more like 4,5,6. I would be able to do it
> only when I get at a packet with a sequence number above window size. This
> may be highly undesirable if we have large window sizes say 1024, 2048,
> etc. we wouldn't be able to drain any more packets until a packet above
> window size is received. Also since the struct rte_rorder_buffer are
> defined in the source .c file, we are not able to move head of buffer from
> applications. *Is there any specific reasons why it is implemented like
> this in the library. Are there any ways to overcome this.*
> --
> Thanks and Regards
> Suraj R Gupta
>


-- 
Thanks and Regards
Suraj R Gupta

Re: [dpdk-dev] [dpdk-users] Issue in draining packets when head is null

[Muhammad Zain-ul-Abideen]

I see the current implementation doesn't cater for this escape sequence. U
have to implement a little logic to traverse through the empty buffs w.r.t.
current packets in buffer via some count.

On Thu, Jan 23, 2020, 10:54 PM Suraj R Gupta <surajrgupta@iith.ac.in> wrote:

> Folks,
> Any Ideas on this.
>
> On Mon, Jan 13, 2020 at 1:04 PM Suraj R Gupta <surajrgupta@iith.ac.in>
> wrote:
>
> > Hi all,
> > This is regarding an issue in the draining packet using
> rte_reorder_drain.
> > Suppose we have a window size of 16 and we get packets with sequence no
> > like 0,1,2,4,5,6. On performing drain operation, I retrieve packets like
> > 0,1,2.
> > Since now order_buf->entries[order_buf->head] i.e. order_buf->entries[3]
> > is null i won't be able to any more like 4,5,6. I would be able to do it
> > only when I get at a packet with a sequence number above window size.
> This
> > may be highly undesirable if we have large window sizes say 1024, 2048,
> > etc. we wouldn't be able to drain any more packets until a packet above
> > window size is received. Also since the struct rte_rorder_buffer are
> > defined in the source .c file, we are not able to move head of buffer
> from
> > applications. *Is there any specific reasons why it is implemented like
> > this in the library. Are there any ways to overcome this.*
> > --
> > Thanks and Regards
> > Suraj R Gupta
> >
>
>
> --
> Thanks and Regards
> Suraj R Gupta
>