Re: [akka-user] remoting comparison: Artery vs Netty

399 views
Skip to first unread message

Patrik Nordwall

unread,
Jul 3, 2017, 7:07:01 AM7/3/17
to akka...@googlegroups.com
What is the unit of the values in the table, e.g. 416? I guess it's milliseconds?

That doesn't match our tests at all. We have measured 800k msg/s for the 100 byte payload.

I'd recommend that you use something bigger than micro instances. Corresponding to 4 cores at least. I don't know how much the far apart regions matters, we have only tried in same zone.

On Mon, Jul 3, 2017 at 11:42 AM, 'Kostiantyn Gorustovych' via Akka User List <akka...@googlegroups.com> wrote:
Hi guys. We are migrating our application from akka 2.4 to 2.5 and among other there is new remoting available.
The throughput Before switching from netty to artery we decided to compare them.

So I created the app with two modules: sender & receiver with 16 remote actors spawned by sender on receiver side.
The finished transmission is that one in which the payload delivered to receiver and message id returned to sender.

The tests implemented as a series of messaging batches. Each batch consisted of 1000 message transmissions of K bytes (per message).
I tested it with messages of 100 byte, 1000 byte, 10000 byte with different idle-cpu-level.
With 100 byte message the artery's throughput looks better than netty's (up to 26%), but for the rest it seems that netty beat artery (40% for 1k message and 60% for 10k message).

Some info about test execution:
- warm up the system with 50 batches of 100 bytes messages;
- measure the average for 100 batches of 100 bytes messages;
- measure the average for 50 batches of 1000 bytes messages;
- measure the average for 115 batches of 10000 bytes messages;

I tested it on AWS micro instances in different regions.

Is it expected results?

idle-cpu-level is represented as Artery X, e.g. Artery 1
AWS micro instances (singapore ↔ frankfurt)
messages / bytes per messageArtery 1Artery 3Artery 5Artery 7Artery 9Artery 10Netty
1k / 100441400416445466969543
1k / 1k1655167116741790175537281217
1k / 10k1400515237191421946320833375398565

--
>>>>>>>>>> Read the docs: http://akka.io/docs/
>>>>>>>>>> Check the FAQ: http://doc.akka.io/docs/akka/current/additional/faq.html
>>>>>>>>>> Search the archives: https://groups.google.com/group/akka-user
---
You received this message because you are subscribed to the Google Groups "Akka User List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to akka-user+unsubscribe@googlegroups.com.
To post to this group, send email to akka...@googlegroups.com.
Visit this group at https://groups.google.com/group/akka-user.
For more options, visit https://groups.google.com/d/optout.



--

Patrik Nordwall
Akka Tech Lead
Lightbend -  Reactive apps on the JVM
Twitter: @patriknw

Viktor Klang

unread,
Jul 3, 2017, 7:54:24 AM7/3/17
to Akka User List
Likely bottlenecked either by network or serialization method (or quite possibly both?).

--
Cheers,
Reply all
Reply to author
Forward
0 new messages