You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@ignite.apache.org by Sambhaji Sawant <sa...@gmail.com> on 2018/06/13 07:54:59 UTC

About Ignite Thin client performance

Hello,
I am getting confused to use Ignite Thin client when I have read following
article about.please let me know is this true


Thin client is up to 50% slower than Ignite client node

Unlike Ignite client node, which knows the latest partition assignments and
gets/puts data right from/to the server node where data is located, the
thin client always communicates to the same Ignite server and then the
server talks to another server where data is located. That means up to
twice more network roundtrips comparing with Ignite client node.
Thus, while the thin client is no slower than Ignite client node on a
"single node" cluster, the more nodes are in the cluster the greater is the
performance difference.

Re: About Ignite Thin client performance

Posted by Denis Magda <dm...@gridgain.com>.
That statement is not generic as Pavel and Igor stated, and I surprised to
see that it ended up in our documentation. Removed it from there.

Sure, it's not a question that the thin client would be slower than a
standard one but that was an incorrect note.

--
Denis

On Wed, Jun 13, 2018 at 6:58 AM Igor Sapego <is...@apache.org> wrote:

> +1 to Pavel
>
> "up to 50%" may mean 0.5% for your specific use case.
> Always measure your use case.
>
> Best Regards,
> Igor
>
>
> On Wed, Jun 13, 2018 at 4:29 PM Pavel Tupitsyn <pt...@apache.org>
> wrote:
>
>> Keep in mind that these performance numbers may be totally irrelevant for
>> your usage patterns and workloads.
>> 50% slowdown can occur in a very simple use case (like cache.get()) in
>> ideal conditions,
>> when there is nothing else but network transfer and deserialization.
>>
>> In real world use cases these network costs may become minuscule compared
>> to the real query and processing times.
>>
>> You should always measure your specific use case and decide.
>>
>> On Wed, Jun 13, 2018 at 11:20 AM, Mikael <mi...@telia.com>
>> wrote:
>>
>>> Hi!
>>>
>>> It's in the documentation so why wouldn't it be true ? you have the same
>>> description at the beginning on how it works:
>>>
>>> "The thin client simply establishes a socket connection to a standard
>>> Ignite node​ and performs all operations through that node."
>>>
>>> Mikael
>>>
>>> Den 2018-06-13 kl. 09:54, skrev Sambhaji Sawant:
>>>
>>> Thin client is up to 50% slower than Ignite client node
>>>
>>>
>>>
>>

Re: About Ignite Thin client performance

Posted by Igor Sapego <is...@apache.org>.
+1 to Pavel

"up to 50%" may mean 0.5% for your specific use case.
Always measure your use case.

Best Regards,
Igor


On Wed, Jun 13, 2018 at 4:29 PM Pavel Tupitsyn <pt...@apache.org> wrote:

> Keep in mind that these performance numbers may be totally irrelevant for
> your usage patterns and workloads.
> 50% slowdown can occur in a very simple use case (like cache.get()) in
> ideal conditions,
> when there is nothing else but network transfer and deserialization.
>
> In real world use cases these network costs may become minuscule compared
> to the real query and processing times.
>
> You should always measure your specific use case and decide.
>
> On Wed, Jun 13, 2018 at 11:20 AM, Mikael <mi...@telia.com>
> wrote:
>
>> Hi!
>>
>> It's in the documentation so why wouldn't it be true ? you have the same
>> description at the beginning on how it works:
>>
>> "The thin client simply establishes a socket connection to a standard
>> Ignite node​ and performs all operations through that node."
>>
>> Mikael
>>
>> Den 2018-06-13 kl. 09:54, skrev Sambhaji Sawant:
>>
>> Thin client is up to 50% slower than Ignite client node
>>
>>
>>
>

Re: About Ignite Thin client performance

Posted by Pavel Tupitsyn <pt...@apache.org>.
Keep in mind that these performance numbers may be totally irrelevant for
your usage patterns and workloads.
50% slowdown can occur in a very simple use case (like cache.get()) in
ideal conditions,
when there is nothing else but network transfer and deserialization.

In real world use cases these network costs may become minuscule compared
to the real query and processing times.

You should always measure your specific use case and decide.

On Wed, Jun 13, 2018 at 11:20 AM, Mikael <mi...@telia.com> wrote:

> Hi!
>
> It's in the documentation so why wouldn't it be true ? you have the same
> description at the beginning on how it works:
>
> "The thin client simply establishes a socket connection to a standard
> Ignite node​ and performs all operations through that node."
>
> Mikael
>
> Den 2018-06-13 kl. 09:54, skrev Sambhaji Sawant:
>
> Thin client is up to 50% slower than Ignite client node
>
>
>

Re: About Ignite Thin client performance

Posted by Mikael <mi...@telia.com>.
Hi!

It's in the documentation so why wouldn't it be true ? you have the same 
description at the beginning on how it works:

"The thin client simply establishes a socket connection to a standard 
Ignite node​ and performs all operations through that node."

Mikael


Den 2018-06-13 kl. 09:54, skrev Sambhaji Sawant:
>
>
>       Thin client is up to 50% slower than Ignite client node
>


Re: About Ignite Thin client performance

Posted by Taehee Kim <sl...@nate.com>.
Hello. I saw the document in Wiki that logic has been improved.
https://cwiki.apache.org/confluence/display/IGNITE/IEP-23%3A+Best+Effort+Affinity+for+thin+clients#IEP23:BestEffortAffinityforthinclients-Tickets



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/

Re: About Ignite Thin client performance

Posted by Taehee Kim <sl...@nate.com>.
Hello, I saw a document Wiki that logic has been improved.
https://cwiki.apache.org/confluence/display/IGNITE/IEP-23%3A+Best+Effort+Affinity+for+thin+clients#IEP23:BestEffortAffinityforthinclients-Tickets



--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/