Comparing different filsystems as the filestore

[Samuel Guo <gu...@gmail.com>]

Hi all,

I am a newbie of Olio. I am now working on evaluating the performance of our
parallel file system dcfs3(a research project) on the web-application
workload. And I found Olio. It seems that Olio is a wonderful benchmark for
web2.0 applications. I plan to compare the performance of the benchmark
while using different filesystems(such as localfs, NFS, and our dcfs3) as
the filestore. but as my limited knowledge about Olio and Web2.0
applications, I have some questions about the benchmark.

1) What is the relationship between concurrent users, registered users, the
storage spaces used in mysql, and the storage spaces used in filestore?
    It seems that more than 50G disk space will be consumed while loading
1000 concurrent users's files.

2) There are a lot components in a Web 2.0 application: Http Servers, MySql
Databases, Memcached instances and file storage. Each component may become
the bottleneck of the system if I didn't configure it right. Is there any
suggestion that how to configure a SUT(System under test) while comparing
the performance of the system using different filesystems as the backend
filestore. Such as, the scale of the test(the number of concurrent users),
the number of nodes used as httpserver, the number of memcached instances,
and so on.
And how to avoid the bottlenecks from other components while comparing
different filesystems as the filestore?

I am looking forward to your reply. And please forgive my limited knowledge
about Olio.

Thanks,
Samuel

Re: Comparing different filsystems as the filestore

[Samuel Guo <gu...@gmail.com>]

Thanks a lot.

On Wed, Mar 18, 2009 at 12:48 AM, Shanti Subramanyam <
Shanti.Subramanyam@sun.com> wrote:

> Samuel Guo wrote:
>
>> 1) What is the relationship between concurrent users, registered users,
>> the
>> storage spaces used in mysql, and the storage spaces used in filestore?
>>    It seems that more than 50G disk space will be consumed while loading
>> 1000 concurrent users's files.
>>
>>
>>
> Actually, it's a lot more than that now. More like 100 GB. The space
> required for the db is still very small - a fraction of a gb (although we
> hope to enhance that). Each concurrent user has 100 users in the db and
> files corresponding to these 100 users in the filestore. See the
> presentation on the olio web page for details on the relationship between
> the various entities. There are two tables in the presentation that gives
> details on storage. http://incubator.apache.org/olio
>

yeah, I found the tables in these presentations. And it is useful for me:)


>
>  2) There are a lot components in a Web 2.0 application: Http Servers,
>> MySql
>> Databases, Memcached instances and file storage. Each component may become
>> the bottleneck of the system if I didn't configure it right. Is there any
>> suggestion that how to configure a SUT(System under test) while comparing
>> the performance of the system using different filesystems as the backend
>> filestore. Such as, the scale of the test(the number of concurrent users),
>> the number of nodes used as httpserver, the number of memcached instances,
>> and so on.
>> And how to avoid the bottlenecks from other components while comparing
>> different filesystems as the filestore?
>>
>>
>>
> That's a tough question. Olio is a macro-level app and workload. As such it
> is difficult to focus on just one component without tuning the whole stack.
> One suggestion I have is that you run the workload at a small scale (say
> 1000 concurrent users) and get a profile of the workload on the filestore.
> If you can get a reasonable understanding of the number of reads/writes/file
> creates etc. you can then write a micro-benchmark to simulate the same load
> on the filesystem without having to worry about tuning the whole Olio stack.
>

It is a good idea. I will try it.


>
> Oops - I really shouldn't be saying this ! I do want more users to come
> join our project :-)


I'd like to join this project and hope that I can contribute something :)


>
>
> If that seems like too much work, you can still stick with Olio. The good
> news is that it does place a substantial stress on the file system (2nd only
> to the web server). Tuning the web server is relatively easy - you should
> find tons of information on apache/php tuning if you google for it. With
> memcached on, the php workload has a minimal impact on the database.
>
>> I am looking forward to your reply. And please forgive my limited
>> knowledge
>> about Olio.
>>
>>
>>
> We all need to start somewhere - welcome to Olio and I hope you find it
> useful.
>
>> Thanks,
>> Samuel
>>
>>
> Shanti
>

Samuel

Re: Comparing different filsystems as the filestore

[Shanti Subramanyam <Sh...@Sun.COM>]

Samuel Guo wrote:
> 1) What is the relationship between concurrent users, registered users, the
> storage spaces used in mysql, and the storage spaces used in filestore?
>     It seems that more than 50G disk space will be consumed while loading
> 1000 concurrent users's files.
>
>   
Actually, it's a lot more than that now. More like 100 GB. The space 
required for the db is still very small - a fraction of a gb (although 
we hope to enhance that). Each concurrent user has 100 users in the db 
and files corresponding to these 100 users in the filestore. See the 
presentation on the olio web page for details on the relationship 
between the various entities. There are two tables in the presentation 
that gives details on storage. http://incubator.apache.org/olio
> 2) There are a lot components in a Web 2.0 application: Http Servers, MySql
> Databases, Memcached instances and file storage. Each component may become
> the bottleneck of the system if I didn't configure it right. Is there any
> suggestion that how to configure a SUT(System under test) while comparing
> the performance of the system using different filesystems as the backend
> filestore. Such as, the scale of the test(the number of concurrent users),
> the number of nodes used as httpserver, the number of memcached instances,
> and so on.
> And how to avoid the bottlenecks from other components while comparing
> different filesystems as the filestore?
>
>   
That's a tough question. Olio is a macro-level app and workload. As such 
it is difficult to focus on just one component without tuning the whole 
stack. One suggestion I have is that you run the workload at a small 
scale (say 1000 concurrent users) and get a profile of the workload on 
the filestore. If you can get a reasonable understanding of the number 
of reads/writes/file creates etc. you can then write a micro-benchmark 
to simulate the same load on the filesystem without having to worry 
about tuning the whole Olio stack.
Oops - I really shouldn't be saying this ! I do want more users to come 
join our project :-)

If that seems like too much work, you can still stick with Olio. The 
good news is that it does place a substantial stress on the file system 
(2nd only to the web server). Tuning the web server is relatively easy - 
you should find tons of information on apache/php tuning if you google 
for it. With memcached on, the php workload has a minimal impact on the 
database.
> I am looking forward to your reply. And please forgive my limited knowledge
> about Olio.
>
>   
We all need to start somewhere - welcome to Olio and I hope you find it 
useful.
> Thanks,
> Samuel
>   
Shanti

Re: Comparing different filsystems as the filestore

[Shanti Subramanyam <Sh...@Sun.COM>]

Samuel Guo wrote:
> 1) What is the relationship between concurrent users, registered users, the
> storage spaces used in mysql, and the storage spaces used in filestore?
>     It seems that more than 50G disk space will be consumed while loading
> 1000 concurrent users's files.
>
>   
Actually, it's a lot more than that now. More like 100 GB. The space 
required for the db is still very small - a fraction of a gb (although 
we hope to enhance that). Each concurrent user has 100 users in the db 
and files corresponding to these 100 users in the filestore. See the 
presentation on the olio web page for details on the relationship 
between the various entities. There are two tables in the presentation 
that gives details on storage. http://incubator.apache.org/olio
> 2) There are a lot components in a Web 2.0 application: Http Servers, MySql
> Databases, Memcached instances and file storage. Each component may become
> the bottleneck of the system if I didn't configure it right. Is there any
> suggestion that how to configure a SUT(System under test) while comparing
> the performance of the system using different filesystems as the backend
> filestore. Such as, the scale of the test(the number of concurrent users),
> the number of nodes used as httpserver, the number of memcached instances,
> and so on.
> And how to avoid the bottlenecks from other components while comparing
> different filesystems as the filestore?
>
>   
That's a tough question. Olio is a macro-level app and workload. As such 
it is difficult to focus on just one component without tuning the whole 
stack. One suggestion I have is that you run the workload at a small 
scale (say 1000 concurrent users) and get a profile of the workload on 
the filestore. If you can get a reasonable understanding of the number 
of reads/writes/file creates etc. you can then write a micro-benchmark 
to simulate the same load on the filesystem without having to worry 
about tuning the whole Olio stack.
Oops - I really shouldn't be saying this ! I do want more users to come 
join our project :-)

If that seems like too much work, you can still stick with Olio. The 
good news is that it does place a substantial stress on the file system 
(2nd only to the web server). Tuning the web server is relatively easy - 
you should find tons of information on apache/php tuning if you google 
for it. With memcached on, the php workload has a minimal impact on the 
database.
> I am looking forward to your reply. And please forgive my limited knowledge
> about Olio.
>
>   
We all need to start somewhere - welcome to Olio and I hope you find it 
useful.
> Thanks,
> Samuel
>   
Shanti