DPDK usage discussions
 help / color / Atom feed
* [dpdk-users] dpdk sw schedule issue
@ 2020-07-30  3:05 Zhiyong Huo
  2020-07-31  8:56 ` Van Haaren, Harry
  0 siblings, 1 reply; 2+ messages in thread
From: Zhiyong Huo @ 2020-07-30  3:05 UTC (permalink / raw)
  To: users

Hi,

We used DPDK to do high performance gateway.
And we chose the event_sw scheduler to do the packet scheduler.
However, there were several issues.

The gateway was set with 32 cores.
I did several test:


1.  One RX polling adapter and  one scheduler. No service core, but 32 cores to  compete the RX adapeter/scheduler adapter.

The performance was too low.



2.  Set multi RX polling adapeters and multi schedulers.

Get higher performance but not as expect.


For the scheduler logic, it seemed that the DPDK has very simple scheduler logic.

Is there any plan or suggestion about the scheduler logic?


Thanks


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [dpdk-users] dpdk sw schedule issue
  2020-07-30  3:05 [dpdk-users] dpdk sw schedule issue Zhiyong Huo
@ 2020-07-31  8:56 ` Van Haaren, Harry
  0 siblings, 0 replies; 2+ messages in thread
From: Van Haaren, Harry @ 2020-07-31  8:56 UTC (permalink / raw)
  To: Zhiyong Huo, users

> -----Original Message-----
> From: users <users-bounces@dpdk.org> On Behalf Of Zhiyong Huo
> Sent: Thursday, July 30, 2020 4:05 AM
> To: users@dpdk.org
> Subject: [dpdk-users] dpdk sw schedule issue
> 
> Hi,

Hi Huo,

> We used DPDK to do high performance gateway.
> And we chose the event_sw scheduler to do the packet scheduler.
> However, there were several issues.

Would you describe your pipeline? Aka, what Eventdev Queues do you configure? (How many, and Atomic or Ordered?)

> The gateway was set with 32 cores.
> I did several test:
> 
> 
> 1.  One RX polling adapter and  one scheduler. No service core, but 32 cores to
> compete the RX adapeter/scheduler adapter.
> The performance was too low.

You mention not using a service core, does this mean that the scheduler is being run
on a thread that is also doing other work? Best performance is often achieved by giving
a dedicated thread for the scheduling work in event_sw.

> 2.  Set multi RX polling adapeters and multi schedulers.
> Get higher performance but not as expect.

The results are not as you expect - do you mean that the packet tx order is not identical to rx?
That would be an expected side-effect of using e.g. RSS to spread work to e.g. 4 NIC rx queues.

Please describe what you are trying to achieve, and what was not as expected in more detail.

> For the scheduler logic, it seemed that the DPDK has very simple scheduler logic.
> Is there any plan or suggestion about the scheduler logic?

The PMD is documented in the DPDK docs here:
https://doc.dpdk.org/guides/eventdevs/sw.html

The scheduling logic is based around Atomic and Ordered packet scheduling logic. I'm
not sure what you mean by very simple scheduler logic.


> Thanks

Regards, -Harry

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, back to index

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-07-30  3:05 [dpdk-users] dpdk sw schedule issue Zhiyong Huo
2020-07-31  8:56 ` Van Haaren, Harry

DPDK usage discussions

Archives are clonable:
	git clone --mirror http://inbox.dpdk.org/users/0 users/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 users users/ http://inbox.dpdk.org/users \
		users@dpdk.org
	public-inbox-index users


Newsgroup available over NNTP:
	nntp://inbox.dpdk.org/inbox.dpdk.users


AGPL code for this site: git clone https://public-inbox.org/ public-inbox