[PATCH v2 1/5] fuse: add fuse node struct

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[PATCH v2 1/5] fuse: add fuse node struct

Jeff
Hi,



I'm trying to learn this patch. But encounter several questions:



1. Do I need to update the LIBFUSE in user space? Where can I find the
related patch for LIBFUSE?



2. I'm interested in the CPU affinity feature. In my LIBFUSE in user space,
there is no affinity to bind the file system process with CPU cores. For
example, we have only 8 processes running on the first 8 cores and waiting
for the request from kernel space on fuse_dev_do_read. If there is a
request from node_id=10, there will be no user space waiting on this "fn"
node, and this request will not be responsed. Anything I missed?



Thanks,

Jeff


------------------------------------------------------------------------------
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH v2 1/5] fuse: add fuse node struct

Ashish Samant
Hi Jeff,

This patch was not merged, so there are no updates to libfuse as well.
Instead we went ahead with another similar approach to solve the
scalability problem. You can find the discussion and git repo for the
patches here:
http://lkml.iu.edu/hypermail/linux/kernel/1509.3/00561.html

Here is the git pull request merged into the 4.2 kernel:
http://lists.openwall.net/linux-kernel/2015/07/02/233

Thanks,
Ashish


On 01/06/2016 10:55 PM, Jeff wrote:

> Hi,
>
>
>
> I'm trying to learn this patch. But encounter several questions:
>
>
>
> 1. Do I need to update the LIBFUSE in user space? Where can I find the
> related patch for LIBFUSE?
>
>
>
> 2. I'm interested in the CPU affinity feature. In my LIBFUSE in user space,
> there is no affinity to bind the file system process with CPU cores. For
> example, we have only 8 processes running on the first 8 cores and waiting
> for the request from kernel space on fuse_dev_do_read. If there is a
> request from node_id=10, there will be no user space waiting on this "fn"
> node, and this request will not be responsed. Anything I missed?
>
>
>
> Thanks,
>
> Jeff
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> fuse-devel mailing list
> [hidden email]
> https://lists.sourceforge.net/lists/listinfo/fuse-devel
>


------------------------------------------------------------------------------
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH v2 1/5] fuse: add fuse node struct

Jeff
Ashish Samant <ashish.samant@...> writes:



>



> Hi Jeff,



>



> This patch was not merged, so there are no updates to libfuse as well.



> Instead we went ahead with another similar approach to solve the



> scalability problem. You can find the discussion and git repo for the



> patches here:



> http://lkml.iu.edu/hypermail/linux/kernel/1509.3/00561.html



>



> Here is the git pull request merged into the 4.2 kernel:



> http://lists.openwall.net/linux-kernel/2015/07/02/233



>



> Thanks,



> Ashish











Hi Ashish







Thanks for your response.







Before I started to explore your new clone approach, I just have a quick



question about CPU affinity feature for you.







>From the application (libfuse) perspective, the file system threads might



be running on any part of the multi-cores. In other words, the threads



(fuse_dev_do_read) might be waiting on part of "fn", but not 100% sure



running on all the cores. While the kernel send a request and is pending on



one of the "fn", it might be not responsed, because there is no user space



thread waiting for this "fn" to provide service. Is it correct? Could you



please let me know what kind of mechasism to address this problem when you



design this patch?











Regards,



Jeff










------------------------------------------------------------------------------
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH v2 1/5] fuse: add fuse node struct

Ashish Samant
Hi Jeff,

On 01/07/2016 04:33 PM, Jeff wrote:

> Ashish Samant <ashish.samant@...> writes:
>> Hi Jeff,
>> This patch was not merged, so there are no updates to libfuse as well.
>> Instead we went ahead with another similar approach to solve the
>> scalability problem. You can find the discussion and git repo for the
>> patches here:
>> http://lkml.iu.edu/hypermail/linux/kernel/1509.3/00561.html 
>> Here is the git pull request merged into the 4.2 kernel:
>> http://lists.openwall.net/linux-kernel/2015/07/02/233 
>> Thanks,
>> Ashish
> Hi Ashish Thanks for your response. Before I started to explore your
> new clone approach, I just have a quick question about CPU affinity
> feature for you. >From the application (libfuse) perspective, the file
> system threads might be running on any part of the multi-cores. In
> other words, the threads (fuse_dev_do_read) might be waiting on part
> of "fn", but not 100% sure running on all the cores. While the kernel
> send a request and is pending on one of the "fn", it might be not
> responsed, because there is no user space thread waiting for this "fn"
> to provide service. Is it correct? Could you please let me know what
> kind of mechasism to address this problem when you design this patch?
This patchset for the kernel was designed alongwith a patchset for
libfuse as well. The libfuse patches create a thread per node (numa or
cpu depending on the mount option) that waits on /dev/fuse
to serve requests on that particular node.   I was waiting for the
kernel pathces to be reviewed before sending the libfuse pathces on the
list. But that does not matter now since we went ahead with a different
approach.
> Regards, Jeff
Thanks,
Ashish


------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel