DragonFly BSD
DragonFly kernel List (threaded) for 2011-03
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: A question on gsoc2011 schedule and how the mentoring works.


From: "Samuel J. Greear" <sjg@xxxxxxxxxxxx>
Date: Tue, 29 Mar 2011 13:05:08 -0600

On Tue, Mar 29, 2011 at 12:38 PM, Nohhyun Park <parkx408@gmail.com> wrote:
> Yes, I have thought about the issue (though not for the reasons you mentioned).
> In fact I have already included in proposal that LIRS should be
> implemented first. (Not because of patent but simply because Linux
> kernel already implements it.)
> Would implementing ARC be a problem for opensource projects as well?
> (Content based chunking using Rabin's fingerprint is also patented but
> people use it in opensource project all the time.)
>
> I was thinking to make the vnode list management as part of mount time
> option and allow different algorithms to be used.
> If the LIRS is the only target than maybe I should just extend current
> LRU implementation since LIRS should always out perform LRU.
>
> So I guess my questions is;
> 1. Is there any value in making vnode list management as mount time option?
> 2. What about algorithms like CLOCK-pro
> (http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.62.377) which
> was used before LIRS and can out perform LIRS at some algorithmic
> overhead?
> 3. Perhap the most important questions is (I have asked this one
> before but no one replied), how critical is vnode management in terms
> of overall performance? Any specific application domain where this
> would become the bottleneck?
>
> I would still like to implement ARC just to compare it's performance
> with other algorithm. (I am a graduate student and I need to convince
> my advisor that this project will result in a paper or two)
> Thank you.
>
> -Nohhyun
>
> -Nohhyun
>
>
>
> On Tue, Mar 29, 2011 at 9:47 AM, Samuel J. Greear <sjg@evilcode.net> wrote:
>> On Tue, Mar 29, 2011 at 8:39 AM, Alex Hornung <ahornung@gmail.com> wrote:
>>> On 29/03/11 07:02, Nohhyun Park wrote:
>>>> Hi.
>>>>  I am a student hoping to work on the "Implement ARC algorithm
>>>> extension for the vnode free list" project.
>>>
>>> Have you given this some thought? ARC is a patented algorithm and I'm
>>> not sure it's the greatest idea to implement[1], knowingly, a patented
>>> algorithm.
>>>
>>> I think you should discuss your thoughts here on this list more before
>>> putting your application forward. Maybe someone else can provide some
>>> other algorithm with similar performance to ARC, yet not patent encumbered.
>>>
>>> Kind Regards,
>>> Alex Hornung
>>>
>>> [1]: http://www.varlena.com/GeneralBits/96.php
>>>
>>
>> LIRS -> http://www.cse.ohio-state.edu/hpcs/WWW/HTML/publications/papers/TR-02-6.pdf
>> was suggested as a possible alternative on IRC.
>>
>> Best,
>> Sam
>>
>>
>
>
>
> --
> -Nohhyun
>
>

Nohhyun,

I think the consensus is that you can absolutely and definitely
implement ARC, just that we probably can't ship it as default so it
would have to be one of a couple of implemented algorithms.

Sam




[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]