Kylin in a multitenant environment

[Victor Sauermann <vi...@gmail.com>]

Hi there!

I'm working as an architect for a german company and we are evaluating the
kylin system for extreme scale OLAP scenarios.
Currently we got our hands around the system that we basically understand
the usage of it. But now it is going further into a multitenant use.

Let me describe the core issue first:

We are an individual software vendor and our software currently resides on
the PCs of our customers. We are planning to migrate these software
products into our own already existing data center. Our portfolio with
nearly 300 applications address various business domains. One main asset is
the ability to drive analytical processes over the data from all of these
applications - and this for example based on OLAP Analytics.
Right now we have more than 40.000 installation sites with a relational
database store. If we migrate this data to our data center we are
definitely facing Big Data solutions. Kylin apparently fits our need. But -
and this is mission critical - each customer needs a dedicated access to
these analytical systems and may use BI Tools like Tableau and such. We
will provide the access to the Kylin system and cope with the needed
authentication. If now the complete data relevant for BI analyitcs reside
in the data cluster for kylin the data shall only be used by its dedicated
responsible customer - each customer must not see in that way data from
other customers.

In facts: Minmal 40.000 customers with in future 4000 parallel accesses.

Our first investigation provides the information, that we have to provide a
dedicated cube for each tenant. Is that correct? This means we need to
define for each interested customer a cube based on the residing data. Will
this cube definition be enough to seperate each tenant from other's data?
Will the system provide enough power to yield that many cube definitions?

Thanks upfront
Victor Sauermann

Re: Kylin in a multitenant environment

[Victor Sauermann <vi...@gmail.com>]

Thanks
We check if we can use this way...
Vic

On Dec 19, 2016 04:26, "Li Yang" <li...@apache.org> wrote:

> One cube can serve multiple customers if all the customers share the same
> model.
>
> 1) Add a "customer" (or "tenant") mandatory dimension in cube
> 2) Make sure all queries take a "customer = xxx" where condition
> 3) Hide SQL interface to your customer and enforce 2)
>
>
> Yang
>
> On Fri, Dec 16, 2016 at 3:16 PM, ShaoFeng Shi <sh...@apache.org>
> wrote:
>
>> Hi Victor,
>>
>> Are these 40.000 customers sharing the same data/cube model? If the model
>> is the same, creating 1 cube will be good for maintenance; creating
>> thousands of cube will be horrible.
>>
>> The open source Kylin's security control is only at Cube level; To get
>> finer control, you need customize or try its enterprise version called KAP.
>>
>> 2016-12-16 14:12 GMT+08:00 Victor Sauermann <vi...@gmail.com>:
>>
>>> Hi there!
>>>
>>> I'm working as an architect for a german company and we are evaluating
>>> the kylin system for extreme scale OLAP scenarios.
>>> Currently we got our hands around the system that we basically
>>> understand the usage of it. But now it is going further into a multitenant
>>> use.
>>>
>>> Let me describe the core issue first:
>>>
>>> We are an individual software vendor and our software currently resides
>>> on the PCs of our customers. We are planning to migrate these software
>>> products into our own already existing data center. Our portfolio with
>>> nearly 300 applications address various business domains. One main asset is
>>> the ability to drive analytical processes over the data from all of these
>>> applications - and this for example based on OLAP Analytics.
>>> Right now we have more than 40.000 installation sites with a relational
>>> database store. If we migrate this data to our data center we are
>>> definitely facing Big Data solutions. Kylin apparently fits our need. But -
>>> and this is mission critical - each customer needs a dedicated access to
>>> these analytical systems and may use BI Tools like Tableau and such. We
>>> will provide the access to the Kylin system and cope with the needed
>>> authentication. If now the complete data relevant for BI analyitcs reside
>>> in the data cluster for kylin the data shall only be used by its dedicated
>>> responsible customer - each customer must not see in that way data from
>>> other customers.
>>>
>>> In facts: Minmal 40.000 customers with in future 4000 parallel accesses.
>>>
>>> Our first investigation provides the information, that we have to
>>> provide a dedicated cube for each tenant. Is that correct? This means we
>>> need to define for each interested customer a cube based on the residing
>>> data. Will this cube definition be enough to seperate each tenant from
>>> other's data? Will the system provide enough power to yield that many cube
>>> definitions?
>>>
>>> Thanks upfront
>>> Victor Sauermann
>>>
>>>
>>
>>
>> --
>> Best regards,
>>
>> Shaofeng Shi 史少锋
>>
>>
>

Re: Kylin in a multitenant environment

[Li Yang <li...@apache.org>]

One cube can serve multiple customers if all the customers share the same
model.

1) Add a "customer" (or "tenant") mandatory dimension in cube
2) Make sure all queries take a "customer = xxx" where condition
3) Hide SQL interface to your customer and enforce 2)


Yang

On Fri, Dec 16, 2016 at 3:16 PM, ShaoFeng Shi <sh...@apache.org>
wrote:

> Hi Victor,
>
> Are these 40.000 customers sharing the same data/cube model? If the model
> is the same, creating 1 cube will be good for maintenance; creating
> thousands of cube will be horrible.
>
> The open source Kylin's security control is only at Cube level; To get
> finer control, you need customize or try its enterprise version called KAP.
>
> 2016-12-16 14:12 GMT+08:00 Victor Sauermann <vi...@gmail.com>:
>
>> Hi there!
>>
>> I'm working as an architect for a german company and we are evaluating
>> the kylin system for extreme scale OLAP scenarios.
>> Currently we got our hands around the system that we basically understand
>> the usage of it. But now it is going further into a multitenant use.
>>
>> Let me describe the core issue first:
>>
>> We are an individual software vendor and our software currently resides
>> on the PCs of our customers. We are planning to migrate these software
>> products into our own already existing data center. Our portfolio with
>> nearly 300 applications address various business domains. One main asset is
>> the ability to drive analytical processes over the data from all of these
>> applications - and this for example based on OLAP Analytics.
>> Right now we have more than 40.000 installation sites with a relational
>> database store. If we migrate this data to our data center we are
>> definitely facing Big Data solutions. Kylin apparently fits our need. But -
>> and this is mission critical - each customer needs a dedicated access to
>> these analytical systems and may use BI Tools like Tableau and such. We
>> will provide the access to the Kylin system and cope with the needed
>> authentication. If now the complete data relevant for BI analyitcs reside
>> in the data cluster for kylin the data shall only be used by its dedicated
>> responsible customer - each customer must not see in that way data from
>> other customers.
>>
>> In facts: Minmal 40.000 customers with in future 4000 parallel accesses.
>>
>> Our first investigation provides the information, that we have to provide
>> a dedicated cube for each tenant. Is that correct? This means we need to
>> define for each interested customer a cube based on the residing data. Will
>> this cube definition be enough to seperate each tenant from other's data?
>> Will the system provide enough power to yield that many cube definitions?
>>
>> Thanks upfront
>> Victor Sauermann
>>
>>
>
>
> --
> Best regards,
>
> Shaofeng Shi 史少锋
>
>

Re: Kylin in a multitenant environment

[ShaoFeng Shi <sh...@apache.org>]

Hi Victor,

Are these 40.000 customers sharing the same data/cube model? If the model
is the same, creating 1 cube will be good for maintenance; creating
thousands of cube will be horrible.

The open source Kylin's security control is only at Cube level; To get
finer control, you need customize or try its enterprise version called KAP.

2016-12-16 14:12 GMT+08:00 Victor Sauermann <vi...@gmail.com>:

> Hi there!
>
> I'm working as an architect for a german company and we are evaluating the
> kylin system for extreme scale OLAP scenarios.
> Currently we got our hands around the system that we basically understand
> the usage of it. But now it is going further into a multitenant use.
>
> Let me describe the core issue first:
>
> We are an individual software vendor and our software currently resides on
> the PCs of our customers. We are planning to migrate these software
> products into our own already existing data center. Our portfolio with
> nearly 300 applications address various business domains. One main asset is
> the ability to drive analytical processes over the data from all of these
> applications - and this for example based on OLAP Analytics.
> Right now we have more than 40.000 installation sites with a relational
> database store. If we migrate this data to our data center we are
> definitely facing Big Data solutions. Kylin apparently fits our need. But -
> and this is mission critical - each customer needs a dedicated access to
> these analytical systems and may use BI Tools like Tableau and such. We
> will provide the access to the Kylin system and cope with the needed
> authentication. If now the complete data relevant for BI analyitcs reside
> in the data cluster for kylin the data shall only be used by its dedicated
> responsible customer - each customer must not see in that way data from
> other customers.
>
> In facts: Minmal 40.000 customers with in future 4000 parallel accesses.
>
> Our first investigation provides the information, that we have to provide
> a dedicated cube for each tenant. Is that correct? This means we need to
> define for each interested customer a cube based on the residing data. Will
> this cube definition be enough to seperate each tenant from other's data?
> Will the system provide enough power to yield that many cube definitions?
>
> Thanks upfront
> Victor Sauermann
>
>


-- 
Best regards,

Shaofeng Shi 史少锋