This is purely speculative, so please don't think I know how to do this. I'm throwing this out so hopefully, if a good idea, the right people can create the example. Memcached has a good assortment of Perl clients. Nginx has a means of embedding Perl into the configuration. Couldn't there be a way of combining these to immediately populate the cache when the cache returns missing? By the way, I'm looking at this as a means of improving I/O for static pages (on SliceHost). Ideally nginx's Memcached Module would have the ability to do this when the requested filename exists on the hard drive: (-f $request_filename/index.html) Daniel Rhoden Director of Information Technology Interactive Internet Websites, Inc. 3854 - 2 Killearn Court Tallahassee, Florida 32309 Voice: (256) 878-5554 E-Mail: firstname.lastname@example.org Website: iiwinc.com
on 2009-03-04 23:31
on 2009-03-05 11:49
Rhoden, Yes it can be done and i also thought about doing this but eventually did not wanted perl-enabled nginx as the main proxy server. If you want an example, I can post it on the wiki. It will be similar to this one. Basically. Check if the file exists If yes, return file if not, call a backend url return response best regards
on 2009-03-05 11:53
Daniel, I see now your other posts, I do not think that you will get any benifit of using memcached with static files. Nginx is already very optimized at serving static files.
on 2009-03-05 12:44
When I tested using memcached vs files on the filesystem, there wasn't much difference between them, and sometimes using memcached fared worse than directly sending the files from the filesystem (this would probably vary quite a bit, depending on the system). The reason for this is that the overhead of creating a request to a memcached server, sending the request, waiting for the response (which is done asynchronously so as not to block the server), then handling the response is 'relatively' high. If the memory cache was instead integrated into Nginx, rather than using an external process like memcached, then it would almost certainly be quite a bit quicker. As far as I know, there isn't an internal memory-cache module for Nginx yet (I'm not sure if ncache has included memory caching yet, or if they intend on doing so). As Atif says, there is little benefit to using memcache for serving static files, but there definitely is for dynamically-generated files. As for Perl, I believe the current implementation is still blocking (i.e. you can't currently generate Perl in the background whilst still serving static files, you can either do one or the other). This would kill the performance terribly. For populating the cache, have you looked at the error_file directive. You can set the error file to script which could be passed to an FCGI process, so that on a cache-miss the script is called to generate the page, which could in the process put the file into the memcached cache. This of course only really makes sense for dynamically-generate content. Marcus.
on 2009-03-05 16:59
On Thu, Mar 5, 2009 at 12:34 PM, Marcus Clyne <email@example.com> wrote: > > For populating the cache, have you looked at the error_file directive. You > can set the error file to script which could be passed to an FCGI process, > so that on a cache-miss the script is called to generate the page, which > could in the process put the file into the memcached cache. > Marcus, I believe Daniel was talking about the opposite. He want to populate the cache if nginx hits the file instead of when it does not find it. So something like this (this is how nginx would do it request comes from /files/1.txt 1. Check if memcache exists and serve from there. 2. Check if file exist and serve from here. <---- This is where he wants to hook it :-) 3. If file does not exists, handle error. What Daniel wants (if I understood correctly) request comes from /files/1.txt 1. Check if memcache exists and serve from there. 2. if(-f file ) call the perl/fcgi process that populates the memcache cache and returns the file. (so this would be done only where there is no cache in memcache) 3. If file does not exists, handle error. Still I dont see whats the point of this but yes it is doable. best regards
on 2009-03-05 20:19
Atif, I understood it the same way you did, and was saying that serving files from memcached as opposed to the filesystem has negligible benefits, and sometimes actually performs worse. The reference to error_page (not error_file as I said before - memory slip) was just as a means to populate memcache when there's no data in the cache - you could easily use this method to put the file into memcache from the filesystem. You wouldn't need to use an Nginx module to put the file into memcache (at least not with the current Perl module - it's blocking, so would slow everything down too much, though with a non-blocking version it could be ok) - though I understand that that's what he's talking about and where he's talking about hooking it in. Another method would be to use the try_files directive. Overall, though, I can only see populating memcache with a file to serve through Nginx as being overall slower than serving the file from the filesystem. In my tests, memcache vs filesystem is pretty similar, and adding any kind of overhead to put a file in memcache will mean just serving the file statically will be more efficient (and won't waste memory unnecessarily). I could see storing static files in an internal memory cache being a bit quicker than serving files statically, but actually not all that much, and would only possibly be of real benefit for files that are under very high demand (e.g. a homepage or a logo on a high-traffic site). Cheers, Marcus. Given that the
on 2009-03-05 21:46
There has been a growing trend in using VPS like Xen to host sites. What I'm seeing is that memory is guaranteed, but the I/O to the hard drive is competing with the other customers on the same physical server. I've given up on the idea of using Perl to populate memcached on the fly for anything, static html or dynamic. I agree that the best place for memory caching of static content is right with nginx. Did some analysis and found that about 1% of our pages get 40% of the traffic. Since these are static pages it is unlikely that they will change in popularity from day to day. So I've decided that a cron job can populate the cache for these few pages (and then some) with very little overhead, very little memory, with the biggest payoff. Thank you all for your feedback on this idea. By the way: I am seeing a boost in performance by caching static files. Small, but every ms counts.
on 2009-03-05 22:58
Daniel, I see this as a good compromise for your situation.
on 2009-03-05 23:00
Marcus, I agree with you. It was a very strange scenario.
on 2009-03-06 06:16
You might want to check out the following if you have not seen it already. http://www.igvita.com/2008/02/11/nginx-and-memcach...
on 2009-03-06 08:25
Daniel, A cron job sounds like a good idea to me too. As part of a project I'm working on, I'll be developing an in-memory cache for Nginx. I'll let you know when it's stable, in case you'd like to try it out. Wrt VPS's, have you looked at Open VZ (i.e. Virtuozzo) VPS ISPs? I'm currently using Tektonic, and have generally been happy with the speed of their system. For your money, you will typically get more memory too. The reason is to do with the platform, I believe. I think that Xen doesn't allow dynamic changing of memory size, but VZ does - so you can 'burst' your memory to your needs on VZ, whilst still having a minimum level. I know that Open VZ does have some problems that I don't think Xen does, though (e.g. with memory-mapped files, meaning that Varnish Cache can't - currently - work on Open VZ VPSs). The difference in architecture between Xen and Open VZ might mean that you'd get a better IO performance on that (I think I was getting more than 5000 req/s serving static files on my $15/mo VPS). Just something you might want to look at. Marcus.