Igniters,
As I understand "javadevmtl" prepared some code.
May be someone experienced could run it under java profiler and may be it
will give some useful information where is the slowdown?
--
Alexey Kuznetsov
GridGain Systems
www.gridgain.com
Sergi,
Can you please clarify what reasons for performance degradation you meant?
Does it worth additional investigation or this is a known issue?
-Val
On Wed, Sep 16, 2015 at 7:42 PM, Dmitriy Setrakyan
wrote:
>
>
> On Thu, Sep 17, 2015 at 2:00 AM, Sergi Vladykin
> wrote:
>
>> Ok, I see. Yes,
On Thu, Sep 17, 2015 at 2:58 AM, Alexey Kuznetsov
wrote:
> If it is not snapsohtable there could be at least one use-case: populated
> once and never changed cache.
> For example for some analytics.
> User populate 100500 entries into cache, build index and run 100500
> queries.
>
>
I agree with
On Thu, Sep 17, 2015 at 2:00 AM, Sergi Vladykin
wrote:
> Ok, I see. Yes, I can name at least few things in off-heap
> sql indexes that make things slower and less scalable
> than on-heap ones, but I'm not sure when we will be
> able to improve them.
>
I actually do not see why the performance is
The docs do mention quite profusely O(1) he has a bit of info regarding
performance of the various index types it supports it also support
concurrency. BUT it is dn fast :)
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Btw-anyone-heard-of-CQ-engine-tp1421p1434.
Alexey,
patches are welcome!;)
Sergi
On Sep 17, 2015 2:58 AM, "Alexey Kuznetsov" wrote:
> If it is not snapsohtable there could be at least one use-case: populated
> once and never changed cache.
> For example for some analytics.
> User populate 100500 entries into cache, build index and run 100
Kevin, did you look
at
org.apache.ignite.examples.datagrid.store.hibernate.CacheHibernatePersonStore?
May be it will help you to grab som ideas for your implementation.
On Thu, Sep 17, 2015 at 3:25 AM, kevin wrote:
> I'm trying to implement a CacheStore interface using Hibernate.
> Ie. CacheSto
If it is not snapsohtable there could be at least one use-case: populated
once and never changed cache.
For example for some analytics.
User populate 100500 entries into cache, build index and run 100500 queries.
On Thu, Sep 17, 2015 at 6:27 AM, Sergi Vladykin
wrote:
> Thanks for the pointers!
>
Another option is to put Ignite in some common module for all the
applications that require Ignite. This way Ignite
classes must be loaded only once.
See https://docs.jboss.org/author/display/WFLY9/Class+Loading+in+WildFly
Sergi
2015-09-17 2:27 GMT+03:00 vkulichenko :
> roar109 wrote
> > If I un
javadevmtl wrote
> Val are you talking about me? I thought I was. Unless my mobile phone did
> something funny. I have been posting for a while...
No, this was for Mike. Sorry for the confusion :)
-Val
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/How-to-impl
roar109 wrote
> If I understand your suggestion correctly is to make single Ignite
> instance per jboss in this case to avoid this sync mechanism?
This should help too. But it sounds like that in your case you can simply
disable shared memory communication. To do this add the following in your
con
Thanks for the pointers!
The main difficulty at Ignite side is that we need
index data structure to be snapshotable with O(1) complexity and of course
without copying of the whole thing, so I'm not sure if we will be able
to use something from there as is.
But nevertheless CQEngine at least deserve
Ok, I see. Yes, I can name at least few things in off-heap
sql indexes that make things slower and less scalable
than on-heap ones, but I'm not sure when we will be
able to improve them.
Sergi
2015-09-16 21:35 GMT+03:00 javadevmtl :
> Sergy, point taken, but what i'm trying to point out is the f
javadevmtl wrote
> Yeah it's perforamance is crazy. I hadn't experienced any "buggy" issues
> when i tried. it also has off-heap otion now also. But I find the concept
> of O(1) indexes very interesting...
I think that O(1) is achievable only for hash indexes and therefore can be
used only for equ
Yeah it's perforamance is crazy. I hadn't experienced any "buggy" issues when
i tried. it also has off-heap otion now also. But I find the concept of O(1)
indexes very interesting...
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Btw-anyone-heard-of-CQ-engine-tp
I'm trying to implement a CacheStore interface using Hibernate.
Ie. CacheStore.load() and CacheStore.write() is implemented by calling
Hibernate
A problem I am seeing is that there are some hooks in which fields get set
after an object is written to the datagrid cache (such as the "locking"
number
Tried to use it a couple of years ago as a local (dev-only) alternative to my
data grid based framework: just plugging it instead of the Hazelcast or
Ignite. Superfast, but extremely buggy. I suspect not thread-safe. Haven't
followed it since then. It'd be great if data grids used something like th
It's O(1) index performance. Was wondering if it's something that can be
pluged in to ignite to replace H2 indexes :)
https://code.google.com/p/cqengine/#Applications
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Btw-anyone-heard-of-CQ-engine-tp1421.html
Se
If I understand your suggestion correctly is to make single Ignite instance
per jboss in this case to avoid this sync mechanism?
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/OverlappingFileLockException-in-nodes-startup-same-time-in-Linux-tp1415p1420.html
Sent
Sergy, point taken, but what i'm trying to point out is the following...
In on-heap mode I can easily insert and query 0-15,000,000 without any
performance loss. And if anything even more then 15 million entries in the
cache.
The off-heap cache though instantly within a few thousand entries it's
Regardless of whether it is off-heap or on-heap, performance of indexes
*will* degrade with cache growth and this is unavoidable.
For better understanding: cache is basically a hash map with O(1) operation
complexity, SQL index is basically an AVL tree which has O(log(N))
operation complexity. So
It seems that jboss starts two instances of Ignite in the same jvm in
different classloaders in parallel.
This way synchronization on IpcSharedMemoryNativeLoader.class does not help
much.
The mutex must be changed to some JVM-wide singleton object or class
instead of Ignite class.
It must be easy
Sorry Dimitry, switching to primitives definitely helped me save on heap
usage, but not performance. The issue I reported with off-heap is still
true, it's alot slower then on-heap and gets slower over time as the cache
grows...
--
View this message in context:
http://apache-ignite-users.70518.
I'm facing an issue with my box when I try to run 2 clients (diferent wars)
in jboss eap 6.3 in Linux.
Details:
- 1 Jboss as server node
- 1 Jboss with 2 wars, each war is a ignite as a client connecting to the
server node
- Ignite 1.3.0
- Jboss EAP 6.3
- Linux CentOS (both)
Steps to reproduce:
So yeah, complete puts I am!
Basically I had switched my models' class name and forgot to do so for the
query so it wasn't serializing back, which made it look like it was allot
faster.
Primitive will definitely save you heap space usage. But won't increase
performance.
--
View this message
Humm maybe not, primitives might not help... More testing... i forgot to
switch my model when setting the config. :( So nothing was being returned...
That why it was faster.
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Is-it-me-or-primitives-as-properties-are-
On Wed, Sep 16, 2015 at 6:02 PM, javadevmtl wrote:
> I think I know what it is...
>
> I was using object types instead of primitive types. I switched all
> possible
> properties of my model from Object types to primitives and off-heap seems
> to
> be performing as you said about 20% slower then o
Actually might be wrong primitives might not help...
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/SQL-Performance-indexing-performance-on-heap-vs-off-heap-tp1352p1411.html
Sent from the Apache Ignite Users mailing list archive at Nabble.com.
Val are you talking about me? I thought I was. Unless my mobile phone did
something funny. I have been posting for a while...
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/How-to-implement-limit-0-to-10-by-Ignite-SQL-tp1392p1410.html
Sent from the Apache Ignite
I think I know what it is...
I was using object types instead of primitive types. I switched all possible
properties of my model from Object types to primitives and off-heap seems to
be performing as you said about 20% slower then on heap, but there no
performance decrease so far...
Let me know i
Ok so here is more details...
Running 1.3.3 on single node single server with 32 cores
I had a MyModel as...
Switched to...
1- My SQL queries # of operations jumped from 18,000/sec to 54,000/sec.
2- Total cache size for 15 Million records went from 20GB on-heap usage to
just about 6.5GB on-
please use this as starting point -
https://cwiki.apache.org/confluence/display/IGNITE/How+to+Contribute
--Yakov
2015-09-16 17:29 GMT+03:00 endian675 :
> I'd be more than happy to contribute. Is there an existing example of this
> kind of pattern in the Ignite code base? I am still getting to
I'd be more than happy to contribute. Is there an existing example of this
kind of pattern in the Ignite code base? I am still getting to grips with
Java (not to mention Git, Ignite, et. al...) so having an example to guide
me is likely to yield a better result.
--
View this message in context
It seems this is not possible now. However, I suppose that we can easily
refactor GridJettyRestHandler to allow custom JSON serializers - just
introduce pluggable interface and set current impl as default. Can you help
us with this and contribute to Ignite?
--Yakov
2015-09-16 16:03 GMT+03:00 endi
So I had my model with Long and Integer primitives and I switched them to
long and int and I got 200% increase in performance... with SQL indexing.
Still using on-heap though.
Is this true? If it's the case that should be documented :P
--
View this message in context:
http://apache-ignite-user
What I am actually trying to achieve is to achieve the @JsonAnySetter
functionality from Jackson within Ignite, either by using Jackson directly
or some other method.
>From some searching it seems that it might be possible to use a custom
PropertySetStrategy inside json-lib in order to get the s
Is it possible to use Jackson as the JSON processor rather than
net.sf.json-lib? If so, how can I configure that?
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Using-Jackson-as-JSON-processor-tp1402.html
Sent from the Apache Ignite Users mailing list archive a
37 matches
Mail list logo