You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by DE VITO Dominique <do...@thalesgroup.com> on 2013/05/06 19:27:51 UTC
RE: cost estimate about some Cassandra patchs
> De : aaron morton [mailto:aaron@thelastpickle.com]
> Envoyé : dimanche 28 avril 2013 22:54
> À : user@cassandra.apache.org
> Objet : Re: cost estimate about some Cassandra patchs
>
> > Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
>
> Assuming you mean "have the coordinator send multiple row read/write requests in a single message to replicas"
>
> Pretty sure this has been raised as a ticket before but I cannot find one now.
>
> It would be a significant change and I'm not sure how big the benefit is. To send the messages the coordinator places them in a queue, there is little delay sending. Then it waits on them async. So there may be some saving on networking but from the coordinators point of view I think the impact is minimal.
>
> What is your use case?
Use case = rows with rowkey like (folder id, file id)
And operations read/write multiple rows with same folder id => so, it could make sense to have a partitioner putting rows with same "folder id" on the same replicas.
But so far, Cassandra is not able to exploit this locality as batch effect ends at the coordinator node.
So, my question about the cost estimate for patching Cassandra.
The closest (or exactly corresponding to my need ?) JIRA entries I have found so far are:
CASSANDRA-166: Support batch inserts for more than one key at once
https://issues.apache.org/jira/browse/CASSANDRA-166
=> "WON'T FIX" status
CASSANDRA-5034: Refactor to introduce Mutation Container in write path
https://issues.apache.org/jira/browse/CASSANDRA-5034
=> I am not very sure if it's related to my topic
Thanks.
Dominique
>
> Cheers
>
>
> -----------------
> Aaron Morton
> Freelance Cassandra Consultant
> New Zealand
>
> @aaronmorton
> http://www.thelastpickle.com
On 27/04/2013, at 4:04 AM, DE VITO Dominique <do...@thalesgroup.com>> wrote:
Hi,
We are created a new partitioner that groups some rows with **different** row keys on the same replicas.
But neither the batch_mutate, or the multiget_slice are able to take opportunity of this partitioner-defined placement to vectorize/batch communications between the coordinator and the replicas.
Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
Thanks.
Regards,
Dominique
RE: cost estimate about some Cassandra patchs
Posted by DE VITO Dominique <do...@thalesgroup.com>.
> -----Message d'origine-----
> De : aaron morton [mailto:aaron@thelastpickle.com]
> Envoyé : mardi 7 mai 2013 10:22
> À : user@cassandra.apache.org
> Objet : Re: cost estimate about some Cassandra patchs
>
> > Use case = rows with rowkey like (folder id, file id)
> > And operations read/write multiple rows with same folder id => so, it could make sense to have a partitioner putting rows with same "folder id" on the same > replicas.
> The entire row key the thing we use to make the token used to both locate the replicas and place the row in the node. I don't see that changing.
Well, we can't do that, because of secondary indexes on rows.
Only the C* v2 will allow the row design you mention, with secondary index.
So, this row design you mention is a no go for us, with C* 1.1 or 1.2.
> Have you done any performance testing to see if this is a problem?
Unfortunately, we have just some pieces, today, for doing performance testing. We are beginning. But still, I investigate to know if alternative designs are (at least) possible. Because if no alternative design is easy to develop, then there's no need to compare performance.
The lesson I learnt here is that, if I would restart our project from the beginning, I would start a more extensive performance testing project along with business project development. It's a kind of must-have for a NoSQL database.
So, the only tests we have done so far with our FolderPartitioner is with a one machine-cluster.
As expected, due to the more important work of this FolderPartitioner, the CPU is a better higher (~10%), memory and network consumptions are the same than with RP, but I have strange results for I/O (average hard drive), for example, for a write-only test. I don't know why the I/O consumption could be much higher with our FolderPartitioner than with the RP. So, I am questioning my measurement methods, and my C* understanding.
Well, the use of such FolderPartitioner is quite a long way to go...
Regards.
Dominique
> Cheers
>
> -----------------
> Aaron Morton
> Freelance Cassandra Consultant
> New Zealand
>
> @aaronmorton
> http://www.thelastpickle.com
On 7/05/2013, at 5:27 AM, DE VITO Dominique <do...@thalesgroup.com> wrote:
> > De : aaron morton [mailto:aaron@thelastpickle.com]
> > Envoyé : dimanche 28 avril 2013 22:54
> > À : user@cassandra.apache.org
> > Objet : Re: cost estimate about some Cassandra patchs
> >
> > > Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
> >
>
> > Assuming you mean "have the coordinator send multiple row read/write requests in a single message to replicas"
> >
> > Pretty sure this has been raised as a ticket before but I cannot find one now.
> >
> > It would be a significant change and I'm not sure how big the benefit is. To send the messages the coordinator places them in a queue, there is little delay sending. Then it waits on them async. So there may be some saving on networking but from the coordinators point of view I think the impact is minimal.
> >
> > What is your use case?
>
> Use case = rows with rowkey like (folder id, file id)
> And operations read/write multiple rows with same folder id => so, it could make sense to have a partitioner putting rows with same "folder id" on the same replicas.
>
> But so far, Cassandra is not able to exploit this locality as batch effect ends at the coordinator node.
>
> So, my question about the cost estimate for patching Cassandra.
>
> The closest (or exactly corresponding to my need ?) JIRA entries I have found so far are:
>
> CASSANDRA-166: Support batch inserts for more than one key at once
> https://issues.apache.org/jira/browse/CASSANDRA-166
> => "WON'T FIX" status
>
> CASSANDRA-5034: Refactor to introduce Mutation Container in write path
> https://issues.apache.org/jira/browse/CASSANDRA-5034
> => I am not very sure if it's related to my topic
>
> Thanks.
>
> Dominique
>
>
>
> >
> > Cheers
> >
> >
> > -----------------
> > Aaron Morton
> > Freelance Cassandra Consultant
> > New Zealand
> >
> > @aaronmorton
> > http://www.thelastpickle.com
>
> On 27/04/2013, at 4:04 AM, DE VITO Dominique <do...@thalesgroup.com> wrote:
>
>
> Hi,
>
> We are created a new partitioner that groups some rows with **different** row keys on the same replicas.
>
> But neither the batch_mutate, or the multiget_slice are able to take opportunity of this partitioner-defined placement to vectorize/batch communications between the coordinator and the replicas.
>
> Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
>
> Thanks.
>
> Regards,
> Dominique
>
>
Re: cost estimate about some Cassandra patchs
Posted by aaron morton <aa...@thelastpickle.com>.
> Use case = rows with rowkey like (folder id, file id)
> And operations read/write multiple rows with same folder id => so, it could make sense to have a partitioner putting rows with same "folder id" on the same replicas.
The entire row key the thing we use to make the token used to both locate the replicas and place the row in the node. I don't see that changing.
Have you done any performance testing to see if this is a problem?
Cheers
-----------------
Aaron Morton
Freelance Cassandra Consultant
New Zealand
@aaronmorton
http://www.thelastpickle.com
On 7/05/2013, at 5:27 AM, DE VITO Dominique <do...@thalesgroup.com> wrote:
> > De : aaron morton [mailto:aaron@thelastpickle.com]
> > Envoyé : dimanche 28 avril 2013 22:54
> > À : user@cassandra.apache.org
> > Objet : Re: cost estimate about some Cassandra patchs
> >
> > > Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
> >
>
> > Assuming you mean "have the coordinator send multiple row read/write requests in a single message to replicas"
> >
> > Pretty sure this has been raised as a ticket before but I cannot find one now.
> >
> > It would be a significant change and I'm not sure how big the benefit is. To send the messages the coordinator places them in a queue, there is little delay sending. Then it waits on them async. So there may be some saving on networking but from the coordinators point of view I think the impact is minimal.
> >
> > What is your use case?
>
> Use case = rows with rowkey like (folder id, file id)
> And operations read/write multiple rows with same folder id => so, it could make sense to have a partitioner putting rows with same "folder id" on the same replicas.
>
> But so far, Cassandra is not able to exploit this locality as batch effect ends at the coordinator node.
>
> So, my question about the cost estimate for patching Cassandra.
>
> The closest (or exactly corresponding to my need ?) JIRA entries I have found so far are:
>
> CASSANDRA-166: Support batch inserts for more than one key at once
> https://issues.apache.org/jira/browse/CASSANDRA-166
> => "WON'T FIX" status
>
> CASSANDRA-5034: Refactor to introduce Mutation Container in write path
> https://issues.apache.org/jira/browse/CASSANDRA-5034
> => I am not very sure if it's related to my topic
>
> Thanks.
>
> Dominique
>
>
>
> >
> > Cheers
> >
> >
> > -----------------
> > Aaron Morton
> > Freelance Cassandra Consultant
> > New Zealand
> >
> > @aaronmorton
> > http://www.thelastpickle.com
>
> On 27/04/2013, at 4:04 AM, DE VITO Dominique <do...@thalesgroup.com> wrote:
>
>
> Hi,
>
> We are created a new partitioner that groups some rows with **different** row keys on the same replicas.
>
> But neither the batch_mutate, or the multiget_slice are able to take opportunity of this partitioner-defined placement to vectorize/batch communications between the coordinator and the replicas.
>
> Does anyone know enough of the inner working of Cassandra to tell me how much work is needed to patch Cassandra to enable such communication vectorization/batch ?
>
> Thanks.
>
> Regards,
> Dominique
>
>