As this is *operating system caches* we're talking about, you can try
"sync" on unix.

Or reboot.

It might also work to run Neo4j in a container (docker, vagrant) and
restart / resume that container.

On Thu, Nov 20, 2014 at 9:29 AM, 'Curtis Mosters' via Neo4j <
[email protected]> wrote:

> Yeah but the issue is that I have 10 comparisons. Each having let's say
> the 10 same words to search for.
>
> So after the first search Neo4j already knows where they are and is faster
> now. That's the issue I have. Real word is something else, where you would
> change those words. But I don't want to think always about new words to
> search for.
>
> So clearing cache would be way better. IS there really no way to do it?
>
> Also tried* killall -9 java*, but not working.
> Also followed On linux, see: http://linux-mm.org/Drop_Caches
> <http://www.google.com/url?q=http%3A%2F%2Flinux-mm.org%2FDrop_Caches&sa=D&sntz=1&usg=AFQjCNGlvcLH54xciQ5X723RMpzCoWQfBg>
> but nothing happens?
>
> Am Mittwoch, 19. November 2014 22:55:23 UTC+1 schrieb Michael Hunger:
>>
>> But for realistic use-cases you will have exactly that setup, that you
>> have hot data in your OS' file-system and also database caches,
>>
>> that is the state you want to reach for real benchmarks as this
>> represents the real world usage, not the cold caches after a computer and
>> database start.
>>
>> The cold cache numbers can be completely ignored imho as they only
>> measure the speed of the disk and the loading mechanism of the FS and
>> database to get data loaded.
>>
>> Same goes for JVM JIT and other optimizations that happen behind the
>> scenes (by OS, JVM, DB)
>>
>> Michael
>>
>> On Wed, Nov 19, 2014 at 9:37 PM, Jacob Hansson <[email protected]>
>> wrote:
>>
>>> Hey mate,
>>>
>>> yeah, this does clarify it. It is very likely that what is happening is
>>> that the OS is caching the index files in RAM, so the second time you run
>>> the database (even after a restart), it does not have to hit disk. You
>>> could verify that this is the case by evicting the OS page cache between
>>> your benchmarks. What OS are you using?
>>>
>>> On linux, see: http://linux-mm.org/Drop_Caches
>>> On mac, you should be able to run 'purge' to clear the caches, see:
>>> http://www.cnet.com/news/purge-the-os-x-disk-cache-to-
>>> analyze-memory-usage/
>>>
>>> For windows there does not seem to be a vendor-provided mechanism to do
>>> this, see here for alternatives: http://stackoverflow.com/questions/
>>> 7405868/how-to-invalidate-the-file-system-cache
>>>
>>> /jake
>>>
>>> On Wed, Nov 19, 2014 at 12:25 PM, 'Curtis Mosters' via Neo4j <
>>> [email protected]> wrote:
>>>
>>>> Hi Jacob, well yeah here is an example:
>>>>
>>>> START n=node:titles("title:solar") RETURN count(*)
>>>>
>>>> If I do this right after the import. It needs *180 sec*. But doing
>>>> this a second time needs *3 sec* and after that again just needs 1,5
>>>> sec, but after that there is no more improvement. Also after doing some
>>>> stuff in the settings, I never get to the 180 sec again.=/
>>>>
>>>> But if you think now it is just that word. No it's not. Also tested 9
>>>> others and here it was lasting round about 60 sec, too. But searching again
>>>> on them was 1-2 sec.
>>>>
>>>> Also after restarting the Neo4j server it is still that fast (1-3 sec).
>>>> And yeah I need the uncached result time for my comparison. It is not
>>>> correct if I continue now with cached things.
>>>>
>>>> Maybe now it's way clearer? If not just let me know. Thank you.
>>>>
>>>> Am Mittwoch, 19. November 2014 18:35:38 UTC+1 schrieb Jacob Hansson:
>>>>>
>>>>> Curtis,
>>>>>
>>>>> can you clarify what you mean by caching, and how you are determining
>>>>> that things are getting cached?
>>>>>
>>>>> If you are talking about caching of the actual data, note that there
>>>>> are several layers of caching - the OS will cache files in its page cache,
>>>>> for instance. If you want to work around that in your testing you need to
>>>>> ask the operating system to flush its caches.
>>>>>
>>>>> /j
>>>>>
>>>>> On Wed, Nov 19, 2014 at 7:36 AM, 'Curtis Mosters' via Neo4j <
>>>>> [email protected]> wrote:
>>>>>
>>>>>> Well for my Benchmark I need a clear Cache because otherwhise Neo4j
>>>>>> is always caching and faking my results in a bad way.
>>>>>>
>>>>>> So I tried:
>>>>>> cache_type=none
>>>>>> cache.memory_ratio=0.0
>>>>>>
>>>>>>
>>>>>>
>>>>>> in the neo4j.properties. Had no real impact still caching.
>>>>>>
>>>>>> Also found something from Michael with jconsole:
>>>>>> http://stackoverflow.com/questions/26189351/neo4j-server-
>>>>>> clear-the-cache-in-ram
>>>>>> But really no idea what he means. =( I meant the JMX call I found
>>>>>> http://stackoverflow.com/questions/12621963/clear-ehcache-
>>>>>> of-remote-server
>>>>>> But also not sure if it is the right thing to clear.
>>>>>>
>>>>>> Here I found infos about the caching: http://neo4j.com/docs/stable/c
>>>>>> onfiguration-caches.html
>>>>>> But nothing worked or was in my mind useful to test.
>>>>>>
>>>>>> Then I gone through all settings here: http://neo4j.com/docs/stable/k
>>>>>> ernel-configuration.html
>>>>>> But yeah except the two lines above nothing looking satisfying. Even
>>>>>> tried query_cache_size=0 but here I got a funny error message then, so 
>>>>>> yeah
>>>>>> dunno what that setting means.
>>>>>>
>>>>>> Then I was thinking that setting the neo_mapping stuff to 0 might
>>>>>> help. But then I found http://grokbase.com/t/gg/neo4j
>>>>>> /1312y592r4/caching-the-whole-graph which says that setting to 0 is
>>>>>> like limiting to infinity.
>>>>>>
>>>>>> Ohman so nothing worked to disable the cache. Why is it so hard to
>>>>>> give a setting disable cache or clearing the cache after server is shut
>>>>>> down.
>>>>>>
>>>>>> Really need this =/
>>>>>>
>>>>>> Thank you
>>>>>>
>>>>>> --
>>>>>> You received this message because you are subscribed to the Google
>>>>>> Groups "Neo4j" group.
>>>>>> To unsubscribe from this group and stop receiving emails from it,
>>>>>> send an email to [email protected].
>>>>>> For more options, visit https://groups.google.com/d/optout.
>>>>>>
>>>>>
>>>>>  --
>>>> You received this message because you are subscribed to the Google
>>>> Groups "Neo4j" group.
>>>> To unsubscribe from this group and stop receiving emails from it, send
>>>> an email to [email protected].
>>>> For more options, visit https://groups.google.com/d/optout.
>>>>
>>>
>>>  --
>>> You received this message because you are subscribed to the Google
>>> Groups "Neo4j" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to [email protected].
>>> For more options, visit https://groups.google.com/d/optout.
>>>
>>
>>  --
> You received this message because you are subscribed to the Google Groups
> "Neo4j" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>

-- 
You received this message because you are subscribed to the Google Groups 
"Neo4j" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to