On Tue, Apr 16, 2013 at 12:49 AM, Ethan Lim <[email protected]> wrote:
> > Best Regards > Lim Zhi Hao (Ethan) > > > Begin forwarded message: > > *From: *Ethan Lim <[email protected]> > *Subject: **Re: [Monkey] GSOC2013 - Caching Filesystem* > *Date: *April 14, 2013 11:48:59 PM GMT+08:00 > *To: *Eduardo Silva <[email protected]> > > Hi Eduardo, > > I am Ethan, and yeah I have been missing over the weekend as I was > researching on possible caching methods by reading research papers through > my University's access to Computer Science journals. > > 1. While researching, I realized a very interesting idea for web server > caching. It differs significantly from document caching is that studies > shown that people who visit page A would most likely visit page A again. > Furthermore, instead of caching data block, we can cache entire web > document (eg. HTML, CSS, etc). For static content, I would like to propose > a simple, realistic but statistically great method of static caching. For > full details, I have attached the research paper. > > the cache is for entire files. > 2. In overview, this is a novel approach by having a fixed amount of RAM > to store a set of web documents. There exist a time period, Tn, whereby the > server log will be read by C running on FUSE and determine the next set of > documents to cache based on Tn-1. A high cache rate will be achieve based > on past time, which will maximize document cache hit rate based on past > log. The added benefit is that CPU utilization is lesser than complex > efficient algorithms, which is critical for Primary web servers as they > usually have compute task to perform. > > the memory reserved is given when starting up the mount point. the primary goal is to have most accessed files in memory, obviously cache rate mas be implemented. > 3. I would like to seek your opinion on this section of my proposal, > while I continue researching for dynamic content caching. > > it looks like what we are suggesting to implement. Please consider in your proposal to write a stress tool to benchmark the file system. regards, > Best Regards > Lim Zhi Hao (Ethan) > > > On Apr 10, 2013, at 10:30 PM, Eduardo Silva <[email protected]> wrote: > > Hi, > > > On Wed, Apr 10, 2013 at 8:25 AM, Ethan Lim <[email protected]> wrote: > >> Good Morning, >> >> That is great. I do plan to implement it in C though, although I know >> there are many choices. >> >> > C is a must. > > > >> 1. I have another concern, as I understand the project is a generic >> caching FS to all kinds of linux systems. However, there are special >> consideration for Monkey, which if operated on Embedded System, has even >> lessor RAM availability. On the other hand, I am more concerned about the >> higher probability of Cache Misses that may incur more frequently because >> of this limitation. May I propose having two configuration settings for >> users to determine which applications are they using it for, General and >> uPCache (Yeah I gave a name for this FS)? Or a more focused approached for >> embedded web server caching? >> >> > Please keep in mind that this project even can work in embedded, it could > have a better impact in common high production environments. When the file > system is mounted, it will export a unix socket interface to modify > parameters of it internal behavior, like a sysadmin can increase or change > memory size on fly, gather statistics and other stuff. Always in a web > server context, its not only for Monkey, it can be used later by any web > server. > > >> 2. I have read the >> blog<http://edsiper.linuxchile.cl/blog/2013/02/27/architecture-of-a-linux-based-web-server/>. >> It does state why initially Monkey decided not to have a cache because of >> cost. I would like to provide my assistance to that. >> >> > A simple cache plugin is welcome. > > regards, > > > >> Just my two cents. >> >> Best Regards >> Lim Zhi Hao (Ethan) >> >> >> On Apr 10, 2013, at 9:21 PM, Eduardo Silva <[email protected]> wrote: >> >> Hi, >> >> >> On Wed, Apr 10, 2013 at 6:40 AM, Ethan Lim <[email protected]> wrote: >> >>> Good Morning, >>> >>> I am undergraduate computer engineering student and have build my own >>> real-time operating systems as part of a module. I am interested in working >>> of the Caching Filesystem, as it is a very different approach indeed. >>> >>> 1. I have knowledge of basic caching algorithms namely, 2-way set >>> associative, n-way set associative or direct mapped. I would like to know >>> more on any caching algorithms preferred or I could propose my own. >>> >>> Your attention is greatly appreciated. >>> >>> >> The Caching Filesystem project is pretty open to implement different >> mechanism, we are more interested about a cool implementation of this >> proposal more than the "perfect algorithm", it will require a lot of >> hands-on so the proposal must include a description of a layer to replace >> the algorithm easily. If you want to include 3 different algorithms to be >> implemented, is up to you. >> >> Keep in mind that the Caching Filesystem is for Web Servers purposes, >> that means read-only mode and behave better than native FS for web servers >> purposes. >> >> Feel free to send us more questions, >> >> best, >> >> -- >> Eduardo Silva >> http://edsiper.linuxchile.cl >> http://www.monkey-project.com >> >> >> > > > -- > Eduardo Silva > http://edsiper.linuxchile.cl > http://www.monkey-project.com > > > > > _______________________________________________ > Monkey mailing list > [email protected] > http://lists.monkey-project.com/listinfo/monkey > > -- Eduardo Silva http://edsiper.linuxchile.cl http://www.monkey-project.com
_______________________________________________ Monkey mailing list [email protected] http://lists.monkey-project.com/listinfo/monkey
