You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@pig.apache.org by de...@wipro.com on 2011/04/05 15:14:16 UTC
Geographically Distributed Hadoop Cluster
I wish to have a Hadoop cluster with DR (Disaster Recovery), for which I need to have the data backup at a geographically different location. (if there's an eqrthquake or tsunami that hits one location).
I was thinking along these lines - Make Hadoop rack-aware, set a replication factor of 4, and have geographically separated nodes under a single Hadoop namespace, and make sure that atleast one replica gets copied to the other data centre. Basically, I want to have an Active-Active kind of a setup. Is this a feasible idea? If not, why?
Assume that bandwidth is not a limitation - I can have unlimited bandwidth - DR is the primary concern - data just cannot be lost.
If this idea is not good, what is the best way to have an Active-Active DR Hadoop setup?
This question may not be PIG related, but I thought folks here can suggest some good ideas. I am posting the same question on the Hadoop General mailing list also.
Please do not print this email unless it is absolutely necessary.
The information contained in this electronic message and any attachments to this message are intended for the exclusive use of the addressee(s) and may contain proprietary, confidential or privileged information. If you are not the intended recipient, you should not disseminate, distribute or copy this e-mail. Please notify the sender immediately and destroy all copies of this message and any attachments.
WARNING: Computer viruses can be transmitted via email. The recipient should check this email and any attachments for the presence of viruses. The company accepts no liability for any damage caused by any virus transmitted by this email.
www.wipro.com
Re: Geographically Distributed Hadoop Cluster
Posted by Dmitriy Ryaboy <dv...@gmail.com>.
Same thing as Yahoo does.
On Tue, Apr 5, 2011 at 2:07 PM, Alan Gates <ga...@yahoo-inc.com> wrote:
> Same thing as Yahoo or same thing as Deepak suggests?
>
> Alan.
>
>
> On Apr 5, 2011, at 2:05 PM, Dmitriy Ryaboy wrote:
>
> Google apparently does the same with GFS / MR, at least by my reading of
>> the Megastore paper.
>>
>> On Tue, Apr 5, 2011 at 9:59 AM, Alan Gates <ga...@yahoo-inc.com> wrote:
>> The concern I have with that approach is I don't think you can guarantee
>> that Hadoop will never assign tasks to read from the geographically
>> distributed nodes. At Yahoo we have separate Hadoop clusters in separate
>> geographic locations and use tools such as distcp to move data between them.
>>
>> Alan.
>>
>>
>> On Apr 5, 2011, at 6:14 AM, deepak.n85@wipro.com wrote:
>>
>> I wish to have a Hadoop cluster with DR (Disaster Recovery), for which I
>> need to have the data backup at a geographically different location. (if
>> there's an eqrthquake or tsunami that hits one location).
>>
>> I was thinking along these lines - Make Hadoop rack-aware, set a
>> replication factor of 4, and have geographically separated nodes under a
>> single Hadoop namespace, and make sure that atleast one replica gets copied
>> to the other data centre. Basically, I want to have an Active-Active kind of
>> a setup. Is this a feasible idea? If not, why?
>>
>> Assume that bandwidth is not a limitation - I can have unlimited bandwidth
>> - DR is the primary concern - data just cannot be lost.
>>
>> If this idea is not good, what is the best way to have an Active-Active DR
>> Hadoop setup?
>>
>> This question may not be PIG related, but I thought folks here can suggest
>> some good ideas. I am posting the same question on the Hadoop General
>> mailing list also.
>>
>>
>>
>> Please do not print this email unless it is absolutely necessary.
>>
>> The information contained in this electronic message and any attachments
>> to this message are intended for the exclusive use of the addressee(s) and
>> may contain proprietary, confidential or privileged information. If you are
>> not the intended recipient, you should not disseminate, distribute or copy
>> this e-mail. Please notify the sender immediately and destroy all copies of
>> this message and any attachments.
>>
>> WARNING: Computer viruses can be transmitted via email. The recipient
>> should check this email and any attachments for the presence of viruses. The
>> company accepts no liability for any damage caused by any virus transmitted
>> by this email.
>>
>> www.wipro.com
>>
>>
>>
>
Re: Geographically Distributed Hadoop Cluster
Posted by Alan Gates <ga...@yahoo-inc.com>.
Same thing as Yahoo or same thing as Deepak suggests?
Alan.
On Apr 5, 2011, at 2:05 PM, Dmitriy Ryaboy wrote:
> Google apparently does the same with GFS / MR, at least by my
> reading of the Megastore paper.
>
> On Tue, Apr 5, 2011 at 9:59 AM, Alan Gates <ga...@yahoo-inc.com>
> wrote:
> The concern I have with that approach is I don't think you can
> guarantee that Hadoop will never assign tasks to read from the
> geographically distributed nodes. At Yahoo we have separate Hadoop
> clusters in separate geographic locations and use tools such as
> distcp to move data between them.
>
> Alan.
>
>
> On Apr 5, 2011, at 6:14 AM, deepak.n85@wipro.com wrote:
>
> I wish to have a Hadoop cluster with DR (Disaster Recovery), for
> which I need to have the data backup at a geographically different
> location. (if there's an eqrthquake or tsunami that hits one
> location).
>
> I was thinking along these lines - Make Hadoop rack-aware, set a
> replication factor of 4, and have geographically separated nodes
> under a single Hadoop namespace, and make sure that atleast one
> replica gets copied to the other data centre. Basically, I want to
> have an Active-Active kind of a setup. Is this a feasible idea? If
> not, why?
>
> Assume that bandwidth is not a limitation - I can have unlimited
> bandwidth - DR is the primary concern - data just cannot be lost.
>
> If this idea is not good, what is the best way to have an Active-
> Active DR Hadoop setup?
>
> This question may not be PIG related, but I thought folks here can
> suggest some good ideas. I am posting the same question on the
> Hadoop General mailing list also.
>
>
>
> Please do not print this email unless it is absolutely necessary.
>
> The information contained in this electronic message and any
> attachments to this message are intended for the exclusive use of
> the addressee(s) and may contain proprietary, confidential or
> privileged information. If you are not the intended recipient, you
> should not disseminate, distribute or copy this e-mail. Please
> notify the sender immediately and destroy all copies of this message
> and any attachments.
>
> WARNING: Computer viruses can be transmitted via email. The
> recipient should check this email and any attachments for the
> presence of viruses. The company accepts no liability for any damage
> caused by any virus transmitted by this email.
>
> www.wipro.com
>
>
Re: Geographically Distributed Hadoop Cluster
Posted by Dmitriy Ryaboy <dv...@gmail.com>.
Google apparently does the same with GFS / MR, at least by my reading of the
Megastore paper.
On Tue, Apr 5, 2011 at 9:59 AM, Alan Gates <ga...@yahoo-inc.com> wrote:
> The concern I have with that approach is I don't think you can guarantee
> that Hadoop will never assign tasks to read from the geographically
> distributed nodes. At Yahoo we have separate Hadoop clusters in separate
> geographic locations and use tools such as distcp to move data between them.
>
> Alan.
>
>
> On Apr 5, 2011, at 6:14 AM, deepak.n85@wipro.com wrote:
>
> I wish to have a Hadoop cluster with DR (Disaster Recovery), for which I
>> need to have the data backup at a geographically different location. (if
>> there's an eqrthquake or tsunami that hits one location).
>>
>> I was thinking along these lines - Make Hadoop rack-aware, set a
>> replication factor of 4, and have geographically separated nodes under a
>> single Hadoop namespace, and make sure that atleast one replica gets copied
>> to the other data centre. Basically, I want to have an Active-Active kind of
>> a setup. Is this a feasible idea? If not, why?
>>
>> Assume that bandwidth is not a limitation - I can have unlimited bandwidth
>> - DR is the primary concern - data just cannot be lost.
>>
>> If this idea is not good, what is the best way to have an Active-Active DR
>> Hadoop setup?
>>
>> This question may not be PIG related, but I thought folks here can suggest
>> some good ideas. I am posting the same question on the Hadoop General
>> mailing list also.
>>
>>
>>
>> Please do not print this email unless it is absolutely necessary.
>>
>> The information contained in this electronic message and any attachments
>> to this message are intended for the exclusive use of the addressee(s) and
>> may contain proprietary, confidential or privileged information. If you are
>> not the intended recipient, you should not disseminate, distribute or copy
>> this e-mail. Please notify the sender immediately and destroy all copies of
>> this message and any attachments.
>>
>> WARNING: Computer viruses can be transmitted via email. The recipient
>> should check this email and any attachments for the presence of viruses. The
>> company accepts no liability for any damage caused by any virus transmitted
>> by this email.
>>
>> www.wipro.com
>>
>
>
Re: Geographically Distributed Hadoop Cluster
Posted by Alan Gates <ga...@yahoo-inc.com>.
The concern I have with that approach is I don't think you can
guarantee that Hadoop will never assign tasks to read from the
geographically distributed nodes. At Yahoo we have separate Hadoop
clusters in separate geographic locations and use tools such as distcp
to move data between them.
Alan.
On Apr 5, 2011, at 6:14 AM, deepak.n85@wipro.com wrote:
> I wish to have a Hadoop cluster with DR (Disaster Recovery), for
> which I need to have the data backup at a geographically different
> location. (if there's an eqrthquake or tsunami that hits one
> location).
>
> I was thinking along these lines - Make Hadoop rack-aware, set a
> replication factor of 4, and have geographically separated nodes
> under a single Hadoop namespace, and make sure that atleast one
> replica gets copied to the other data centre. Basically, I want to
> have an Active-Active kind of a setup. Is this a feasible idea? If
> not, why?
>
> Assume that bandwidth is not a limitation - I can have unlimited
> bandwidth - DR is the primary concern - data just cannot be lost.
>
> If this idea is not good, what is the best way to have an Active-
> Active DR Hadoop setup?
>
> This question may not be PIG related, but I thought folks here can
> suggest some good ideas. I am posting the same question on the
> Hadoop General mailing list also.
>
>
>
> Please do not print this email unless it is absolutely necessary.
>
> The information contained in this electronic message and any
> attachments to this message are intended for the exclusive use of
> the addressee(s) and may contain proprietary, confidential or
> privileged information. If you are not the intended recipient, you
> should not disseminate, distribute or copy this e-mail. Please
> notify the sender immediately and destroy all copies of this message
> and any attachments.
>
> WARNING: Computer viruses can be transmitted via email. The
> recipient should check this email and any attachments for the
> presence of viruses. The company accepts no liability for any damage
> caused by any virus transmitted by this email.
>
> www.wipro.com