Hi Pavel,


Thanks for the clarification.



Just curious: If CopyOnRead has no effect in .Net, why is it exposed in the
.Net API?



It sounds like if my code was Java, rather than c#, setting CopyOnRead
result in an object reference being passed to the client with no
transmission of data and no deserialization overhead, even if not using
Binary Mode, is that correct? This suggests a pretty significant
performance penalty if you’re using the .Net API and suggests ‘heavy-lift’
aspects of a solution would be much more performant if implemented in Java.
Is that correct?



My reason for asking is that I have workflows that might access 10’s of
thousands of cached items where each cached item could be several Kb in
size and where responses are returned in 200-300 of milliseconds. Some
quick finger in the air tests indicate the deserialization overhead alone
by using a .Net client (at least for those workflows) is much longer than
that.



Thanks,

Raymond.



*From:* Pavel Tupitsyn [mailto:ptupit...@apache.org]
*Sent:* Tuesday, February 14, 2017 2:16 AM
*To:* user@ignite.apache.org
*Subject:* Re: Effective size limit for cache items in Ignite



Hi Raymond,



CopyOnRead setting has no effect in .NET code.

It exists for cases when there are both Java and .NET nodes in a cluster.



In Ignite.NET each cache.Get call causes two things:

1) Copy a piece of memory from Java (serialized data)

2) Deserialize the data into user type instance.



You can avoid deserialization by using Binary Mode:

https://apacheignite-net.readme.io/docs/binary-mode



Pavel



On Mon, Feb 13, 2017 at 6:38 AM, Raymond Wilson <raymond_wil...@trimble.com>
wrote:

I found this code snippet in the Ignite v1.8 source:



This appears to be the core place where the value of CopyOnRead has an
effect, though I don’t understand the context of the other logical
conditions around it; it seems that CopyOnRead is dependent on other
configuration state before it will be honoured. I’m still digging to locate
the detailed help around this setting to understand these other conditions.

    /** {@inheritDoc} */

    @Override public CacheObjectContext contextForCache(CacheConfiguration
ccfg) throws IgniteCheckedException {

        assert ccfg != null;



        CacheMemoryMode memMode = ccfg.getMemoryMode();



        boolean storeVal = !ccfg.isCopyOnRead() || (!isBinaryEnabled(ccfg)
&&

            (GridQueryProcessor.isEnabled(ccfg) ||
ctx.config().isPeerClassLoadingEnabled()));



        CacheObjectContext res = new CacheObjectContext(ctx,

            ccfg.getName(),

            ccfg.getAffinityMapper() != null ? ccfg.getAffinityMapper() :
new GridCacheDefaultAffinityKeyMapper(),

            ccfg.isCopyOnRead() && memMode != OFFHEAP_VALUES,

            storeVal,

            ctx.config().isPeerClassLoadingEnabled() &&
!isBinaryEnabled(ccfg));



        ctx.resource().injectGeneric(res.defaultAffMapper());



        return res;

    }



Thanks,

Raymond.



*From:* Raymond Wilson [mailto:raymond_wil...@trimble.com]
*Sent:* Monday, February 13, 2017 2:43 PM
*To:* user@ignite.apache.org
*Subject:* RE: Effective size limit for cache items in Ignite



Ah, I found the CopyOnRead flag in the cache configuration.



Unfortunately, it seems to have the same behaviour regardless of the
setting for this flag.



If I create an example like the below, it seems that querying the same
element from the cache many times takes about the same amount of time in
both cases. Visual Studio also reports large numbers of GC episodes while
it cleans up the large freed MyCacheClass instances. Is this flag only
applicable to Java contexts? I did also try setting KeepBinaryInStore to
true, though there was no noticeable difference.



[Serializable]

    public class MyCacheClass

    {

        public String name = String.Empty;

        private byte[] localData = null;



        public MyCacheClass(String _name)

        {

            name = _name;

            localData = new byte[4000000];

        }

    }



    class Program

    {

        static void Main(string[] args)

        {

            IIgnite ignite = Ignition.Start();



            // Add a cache to Ignite

            ICache<String, MyCacheClass> cache = ignite.CreateCache<String,
MyCacheClass>

                (new CacheConfiguration()

                {

                    Name = "TestCache",

                    CopyOnRead = false,

                    KeepBinaryInStore = true

                });



            // Add a cache item

            cache.Put("First", new MyCacheClass("FirstItem"));



            // query back the cache items

            for (int i = 0; i < 30000; i++)

            {

                MyCacheClass first = cache.Get("First");

            }

      }





*From:* Raymond Wilson [mailto:raymond_wil...@trimble.com
<raymond_wil...@trimble.com>]
*Sent:* Monday, February 13, 2017 11:35 AM
*To:* user@ignite.apache.org
*Subject:* Effective size limit for cache items in Ignite



Hi,



What is the practical size limit for items in an Ignite cache?



I suspect the answer is something “As large as the memory you have to hold
it”, but my question is more aimed at the practicality of large items in a
cache due to the overhead of pulling copies of the items out of the cache
in response to a Cache.Get() request.



For instance, let’s say I had cache items in the 64Kb size range, and had
requests that commonly refer to those cache items to perform some work on
them in response to a request. Will each Cache.Get() request require an
extraction and repackaging of the cache item prior to handing it back to
the caller as a new (copied) version of that cache item, or is there a way
for just a reference to the cache item to be returned to the caller?



I understand there is a way to designate the information in a cache as just
blobs of data with no serialisation semantics. In this case does a
Cache.Get() return a pointer or a copy (with a local locking semantic to
prevent change)?



Thanks,

Raymond.

Reply via email to