Mailing-List: contact user-help@ignite.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@ignite.apache.org
MIME-Version: 1.0
In-Reply-To: <1445890961848-1710.post@n6.nabble.com>
References: <1445890961848-1710.post@n6.nabble.com>
From: Dmitriy Setrakyan <dsetrakyan@apache.org>
Date: Mon, 26 Oct 2015 18:26:00 -0700
Message-ID: 
 <CA+0=VoVMarOQY4W3POJavDqTPKLm2SeCM6UUmA_p=L6CQkRFcA@mail.gmail.com>
Subject: Re: Performance issue with REPLICATED cache
To: user <user@ignite.apache.org>
Content-Type: multipart/alternative; boundary=001a113d841a36307105230bf8a3

--001a113d841a36307105230bf8a3
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

Hi Yuri,

Is there a chance you can create a small github project reproducing this?
You don=E2=80=99t need to have 15 caches, I think 1 cache would suffice.

If the above is not possible, then we will have to investigate by looking
at logs, code, etc., which would take more time.

D.

On Mon, Oct 26, 2015 at 1:22 PM, Yury Lazouski <yury_lazouski@epam.com>
wrote:

> Hi All,
> we are using Ignite 1.3.3 in win server environment.
>
> *Caches configuration*
>
> Each Ignite node has ~15 caches ~1M records each of Integer-to-Object
> entries (total memory required ~7Gb), configured as "REPLICATED",
> readFromBackup flag is set to "true", transaction mode is "ATOMIC", all t=
he
> rest left defaults. One instance of loader service is configured which
> populates caches using IgniteDataStreamer on regular basis (scheduler) wi=
th
> deltas found in source DB, no other writes/updates are performed on cache=
s,
> from client perspective cache is effectively read-only. For the sake of
> experiment simplification database is getting loaded to caches only once =
-
> no further updates are happening.
>
> *Job execution*
>
> In-process (same jvm) client submits callable task to Ignite node which
> essentially traverses caches via get(key) and does calculations on
> retrieved
> data.
>
> *Problem*
>
> With one Ignite node average get(key) operation time is 20-40 =CE=BCs, wi=
th
> enabling second node timings (same local node timings!) are getting
> increased to >100 =CE=BCs. With adding parallel clients (parallel executi=
on of
> callable tasks which travers same caches) numbers reach millisecond level=
.
> If we configure loader service on each node and switch mode to "LOCAL",
> get(key) time is constant (0-1 =CE=BCs) and performance scalability is li=
near
> with addition of Ignite nodes. So the question is how to make "REPLICATED=
"
> cache perform like "LOCAL" if there are no updates to the caches and data
> is
> only getting accessed locally? Please advise.
>
>
>
> --
> View this message in context:
> http://apache-ignite-users.70518.x6.nabble.com/Performance-issue-with-REP=
LICATED-cache-tp1710.html
> Sent from the Apache Ignite Users mailing list archive at Nabble.com.
>

--001a113d841a36307105230bf8a3
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">Hi Yuri,<div><br></div><div>Is there a chance you can crea=
te a small github project reproducing this? You don=E2=80=99t need to have =
15 caches, I think 1 cache would suffice.</div><div><br></div><div>If the a=
bove is not possible, then we will have to investigate by looking at logs, =
code, etc., which would take more time.</div><div><br></div><div>D.</div></=
div><div class=3D"gmail_extra"><br><div class=3D"gmail_quote">On Mon, Oct 2=
6, 2015 at 1:22 PM, Yury Lazouski <span dir=3D"ltr">&lt;<a href=3D"mailto:y=
ury_lazouski@epam.com" target=3D"_blank">yury_lazouski@epam.com</a>&gt;</sp=
an> wrote:<br><blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;=
border-left:1px #ccc solid;padding-left:1ex">Hi All,<br>
we are using Ignite 1.3.3 in win server environment.<br>
<br>
*Caches configuration*<br>
<br>
Each Ignite node has ~15 caches ~1M records each of Integer-to-Object<br>
entries (total memory required ~7Gb), configured as &quot;REPLICATED&quot;,=
<br>
readFromBackup flag is set to &quot;true&quot;, transaction mode is &quot;A=
TOMIC&quot;, all the<br>
rest left defaults. One instance of loader service is configured which<br>
populates caches using IgniteDataStreamer on regular basis (scheduler) with=
<br>
deltas found in source DB, no other writes/updates are performed on caches,=
<br>
from client perspective cache is effectively read-only. For the sake of<br>
experiment simplification database is getting loaded to caches only once -<=
br>
no further updates are happening.<br>
<br>
*Job execution*<br>
<br>
In-process (same jvm) client submits callable task to Ignite node which<br>
essentially traverses caches via get(key) and does calculations on retrieve=
d<br>
data.<br>
<br>
*Problem*<br>
<br>
With one Ignite node average get(key) operation time is 20-40 =CE=BCs, with=
<br>
enabling second node timings (same local node timings!) are getting<br>
increased to &gt;100 =CE=BCs. With adding parallel clients (parallel execut=
ion of<br>
callable tasks which travers same caches) numbers reach millisecond level.<=
br>
If we configure loader service on each node and switch mode to &quot;LOCAL&=
quot;,<br>
get(key) time is constant (0-1 =CE=BCs) and performance scalability is line=
ar<br>
with addition of Ignite nodes. So the question is how to make &quot;REPLICA=
TED&quot;<br>
cache perform like &quot;LOCAL&quot; if there are no updates to the caches =
and data is<br>
only getting accessed locally? Please advise.<br>
<br>
<br>
<br>
--<br>
View this message in context: <a href=3D"http://apache-ignite-users.70518.x=
6.nabble.com/Performance-issue-with-REPLICATED-cache-tp1710.html" rel=3D"no=
referrer" target=3D"_blank">http://apache-ignite-users.70518.x6.nabble.com/=
Performance-issue-with-REPLICATED-cache-tp1710.html</a><br>
Sent from the Apache Ignite Users mailing list archive at Nabble.com.<br>
</blockquote></div><br></div>

--001a113d841a36307105230bf8a3--