You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hama.apache.org by Claudio Martella <cl...@tis.bz.it> on 2011/06/28 16:01:29 UTC
About the memory requirements of Hama/Pregel
Hi,
I was considering an issue not discussed in Pregel paper. One
interesting thing about MapReduce is that the memory limit for each job
in a cluster node is the number of task assigned to that node * the
memory allocated by each task (plus of course the framework overhead).
The input data is iteratively loaded (and kind of emitted too) from
storage and doesn't need to be kept in memory or loaded together in advance.
In Pregel, to me, it looks like the whole graph has to fit in memory
among the cluster nodes. As it is somehow stateful and doesn't have an
iterative job-job nature as MapReduce, at the beginning of the job all
the vertices and their data have to be loaded into the cluster (and kept
in memory).
So the question is: if i want to run Pagerank over my graph, do I need
to be able to store the whole graph in memory?
Thanks!
--
Claudio Martella
Digital Technologies
Unit Research & Development - Analyst
TIS innovation park
Via Siemens 19 | Siemensstr. 19
39100 Bolzano | 39100 Bozen
Tel. +39 0471 068 123
Fax +39 0471 068 129
claudio.martella@tis.bz.it http://www.tis.bz.it
Short information regarding use of personal data. According to Section 13 of Italian Legislative Decree no. 196 of 30 June 2003, we inform you that we process your personal data in order to fulfil contractual and fiscal obligations and also to send you information regarding our services and events. Your personal data are processed with and without electronic means and by respecting data subjects' rights, fundamental freedoms and dignity, particularly with regard to confidentiality, personal identity and the right to personal data protection. At any time and without formalities you can write an e-mail to privacy@tis.bz.it in order to object the processing of your personal data for the purpose of sending advertising materials and also to exercise the right to access personal data and other rights referred to in Section 7 of Decree 196/2003. The data controller is TIS Techno Innovation Alto Adige, Siemens Street n. 19, Bolzano. You can find the complete information on the web site www.tis.bz.it.
Re: About the memory requirements of Hama/Pregel
Posted by Thomas Jungblut <th...@googlemail.com>.
Hi Claudio,
we don't have an IO System to keep the data, so you're forced to manage this
by yourself. Never the less the example will load your input into a HashMap
for performance. So you're right that you're graph has to fit at least into
RAM.
But at least you can fork your own PageRank example and add disk paging to
it.
Have fun.
2011/6/28 Claudio Martella <cl...@tis.bz.it>
> Hi,
>
> I was considering an issue not discussed in Pregel paper. One
> interesting thing about MapReduce is that the memory limit for each job
> in a cluster node is the number of task assigned to that node * the
> memory allocated by each task (plus of course the framework overhead).
> The input data is iteratively loaded (and kind of emitted too) from
> storage and doesn't need to be kept in memory or loaded together in
> advance.
>
> In Pregel, to me, it looks like the whole graph has to fit in memory
> among the cluster nodes. As it is somehow stateful and doesn't have an
> iterative job-job nature as MapReduce, at the beginning of the job all
> the vertices and their data have to be loaded into the cluster (and kept
> in memory).
>
> So the question is: if i want to run Pagerank over my graph, do I need
> to be able to store the whole graph in memory?
>
>
> Thanks!
>
> --
> Claudio Martella
> Digital Technologies
> Unit Research & Development - Analyst
>
> TIS innovation park
> Via Siemens 19 | Siemensstr. 19
> 39100 Bolzano | 39100 Bozen
> Tel. +39 0471 068 123
> Fax +39 0471 068 129
> claudio.martella@tis.bz.it http://www.tis.bz.it
>
> Short information regarding use of personal data. According to Section 13
> of Italian Legislative Decree no. 196 of 30 June 2003, we inform you that we
> process your personal data in order to fulfil contractual and fiscal
> obligations and also to send you information regarding our services and
> events. Your personal data are processed with and without electronic means
> and by respecting data subjects' rights, fundamental freedoms and dignity,
> particularly with regard to confidentiality, personal identity and the right
> to personal data protection. At any time and without formalities you can
> write an e-mail to privacy@tis.bz.it in order to object the processing of
> your personal data for the purpose of sending advertising materials and also
> to exercise the right to access personal data and other rights referred to
> in Section 7 of Decree 196/2003. The data controller is TIS Techno
> Innovation Alto Adige, Siemens Street n. 19, Bolzano. You can find the
> complete information on the web site www.tis.bz.it.
>
>
>
>
>
--
Thomas Jungblut
Berlin
mobile: 0170-3081070
business: thomas.jungblut@testberichte.de
private: thomas.jungblut@gmail.com
Re: About the memory requirements of Hama/Pregel
Posted by "Edward J. Yoon" <ed...@apache.org>.
In conclusion, I would say yes.
But, I'm not sure that there's another method at the moment.
On Tue, Jun 28, 2011 at 11:01 PM, Claudio Martella
<cl...@tis.bz.it> wrote:
> Hi,
>
> I was considering an issue not discussed in Pregel paper. One
> interesting thing about MapReduce is that the memory limit for each job
> in a cluster node is the number of task assigned to that node * the
> memory allocated by each task (plus of course the framework overhead).
> The input data is iteratively loaded (and kind of emitted too) from
> storage and doesn't need to be kept in memory or loaded together in advance.
>
> In Pregel, to me, it looks like the whole graph has to fit in memory
> among the cluster nodes. As it is somehow stateful and doesn't have an
> iterative job-job nature as MapReduce, at the beginning of the job all
> the vertices and their data have to be loaded into the cluster (and kept
> in memory).
>
> So the question is: if i want to run Pagerank over my graph, do I need
> to be able to store the whole graph in memory?
>
>
> Thanks!
>
> --
> Claudio Martella
> Digital Technologies
> Unit Research & Development - Analyst
>
> TIS innovation park
> Via Siemens 19 | Siemensstr. 19
> 39100 Bolzano | 39100 Bozen
> Tel. +39 0471 068 123
> Fax +39 0471 068 129
> claudio.martella@tis.bz.it http://www.tis.bz.it
>
> Short information regarding use of personal data. According to Section 13 of Italian Legislative Decree no. 196 of 30 June 2003, we inform you that we process your personal data in order to fulfil contractual and fiscal obligations and also to send you information regarding our services and events. Your personal data are processed with and without electronic means and by respecting data subjects' rights, fundamental freedoms and dignity, particularly with regard to confidentiality, personal identity and the right to personal data protection. At any time and without formalities you can write an e-mail to privacy@tis.bz.it in order to object the processing of your personal data for the purpose of sending advertising materials and also to exercise the right to access personal data and other rights referred to in Section 7 of Decree 196/2003. The data controller is TIS Techno Innovation Alto Adige, Siemens Street n. 19, Bolzano. You can find the complete information on the web site www.tis.bz.it.
>
>
>
>
>
--
Best Regards, Edward J. Yoon
@eddieyoon
Re: About the memory requirements of Hama/Pregel
Posted by Paweł Brach <br...@gmail.com>.
2011/6/28 Claudio Martella <cl...@tis.bz.it>
>
>
> So the question is: if i want to run Pagerank over my graph, do I need
> to be able to store the whole graph in memory?
>
>
>
Hi Claudio,
In my opinion you don't need to store the whole graph in memory.
For example, you can store in the HDFS Page Rank records in the following
way:
vertex | adjacency_list
You can split this file into chunks (every BSP node will process exactly one
chunk).
Every Page Rank record could be process separately (there is no need to read
whole chunk into memory).
If you every time read only one PR record and process it it will look like
Map Reduce algorithm, but you can read more PR records (as many as you can -
it depends on the memory constraints).
Best,
Pawel