Page 5 of 6
Re: Scalability
Posted: Tue Jul 07, 2009 4:12 pm
by Christopher
I'm sorry to hear that. Perhaps William or Theory is interested?
Re: Scalability
Posted: Wed Jul 08, 2009 4:35 am
by kaisellgren
If I am to write such articles, someone would need to proofread it; otherwise, it would be monotonic text written in poor English

. Besides, I'm going to be busy soon. Anyway, it seems that not many are interested in this...?
I've been doing some research everyday and I would love to hear others' thoughts.
Re: Scalability
Posted: Wed Jul 08, 2009 4:37 am
by Weirdan
kaisellgren wrote:Anyway, it seems that not many are interested in this...?

How do you know that?
Re: Scalability
Posted: Wed Jul 08, 2009 4:49 am
by kaisellgren
I don't. That's why the word "seems"... it's quite silent here. What about you?
Re: Scalability
Posted: Wed Jul 08, 2009 5:17 am
by SvanteH
That does not mean that people isn't reading what's written here

Re: Scalability
Posted: Wed Jul 08, 2009 5:18 am
by Weirdan
I'm following this topic closely, but have yet to chime in, since it's more about HA setups (and I'm more interested in performance).
Re: Scalability
Posted: Wed Jul 08, 2009 5:22 am
by VladSun
++me

Re: Scalability
Posted: Wed Jul 08, 2009 6:49 am
by kaisellgren
Good to hear. So, does someone want to strike up a discussion? That would probably be in this forum at first and later moved to Tutorials forum as arborint suggested. One question, do we have unlimited time to edit our posts in Devnet?
As what comes to articles, would we have an entire article for one particular solution (Memcached) or for one particular target (Distributing cache and session data among several servers)?
Re: Scalability
Posted: Wed Jul 08, 2009 7:19 am
by Weirdan
I'd vote for per-solutions threads
Re: Scalability
Posted: Wed Jul 08, 2009 9:08 am
by Theory?
arborint wrote:I'm sorry to hear that. Perhaps William or Theory is interested?
I'm not a master by any means. I'd love to help all I can though, so maybe once we agree on a "curriculum" then I can better know where I can help. If nothing else, I'd be an excellent editor

Re: Scalability
Posted: Wed Jul 08, 2009 12:41 pm
by Christopher
I think there is plenty of interest, and different individuals will be more involved in different topics. And perhaps Kai (and others) misunderstood my proposal about the "editorial work." My goal was to find a way to condense the good information spread across an interesting thread (like this one) into a single post/article.
My proposal was that one person would take responsibility for each topic/thread. So Kai might do some of the security topics, Wierdan some of the performance topics, VladSun other topics, etc. The "most interested person" would create the first post and would copy text that others posted in the thread into the first post to create the tutorial. So it is more about gathering and editing text than having to write the whole thing yourself. People could post additions or changes. It would also be possible to split the work. For example, Kai could create the first post which is the general overview information, and William could create the second post that contained Examples/Tutorial. Whatever works. There do not have to be any strict rules. And I am glad to help with the "editorial work."
Take this thread for example. Theory? is the OP. What information spread across all the posts here could he pull into the first post to essentially answer his question? I know pytrin provided many good links and William gave good information on some configurations.
PS - I would be glad to do much of the "editorial work" if people thought the process was worthwhile, but did not have the time for that.
Re: Scalability
Posted: Thu Jul 09, 2009 7:04 am
by wei
squid can be used to achieve higher availability and scale. 2 or 3 squid boxes infront of your web servers can increase your overall capacity, especially if your pages and static content produce the correct cachable headers, e.g. max-age = 30. (i.e. cache pages means we don't even hit apache at all). Squid can be setup to be peered, i.e. shared cache. Can be setup to reply with stale content while re-validating, stale while on error (this is very useful, e.g. if the web server just died and need to switch to another), etc. Squid can be setup on the same boxes as the web server, e.g. squid on port 80, proxying apache on port 81, and then round-robin between multiple of those (e.g. multi-colo failover).
Another usage for squid, stick squid between your web front-end some some backends that are web services.
Re: Scalability
Posted: Thu Jul 09, 2009 9:15 am
by Weirdan
wei wrote:
squid can be used to achieve higher availability and scale.
Or, better yet, nginx. You rarely need such complex proxy server as squid for just reverse-proxying.
Re: Scalability
Posted: Thu Jul 09, 2009 9:29 am
by VladSun
Weirdan wrote:wei wrote:
squid can be used to achieve higher availability and scale.
Or, better yet, nginx. You rarely need such complex proxy server as squid for just reverse-proxying.
I think
wei meant something like this:
http://kevin.vanzonneveld.net/techblog/ ... _1_server/
So, it's not only reverse proxying, it's cache proxying also.
PS: Maybe the terminology is not very clear, at least for me

I mean that when I hear reverse proxy, I expect it to be a simple TCP/IP redirect to another server(s) (e.g. load balancing). No cache mechanisms involved.
Re: Scalability
Posted: Thu Jul 09, 2009 9:57 am
by kaisellgren
Reverse proxies are often used for load balancing a cluster of web servers. In case of Squid, it can often cache most HTTP requests and thus heavy software like Apache are much less needed. Squid is a lightweight caching proxy in front of the web server that can be used for reverse proxy reasons, too. Another nice reverse proxy tool called Pound (
http://www.apsis.ch/pound/) has features like detecting the least overloaded server and use it to serve the content (and does not select servers that are down - fail-over). There's also a software called Varnish (I have never tried,
http://varnish.projects.linpro.no/) which works in a similar way. Maybe I'll do some testing with each of them and compare the results. "Free" performance is always interesting.
arborint wrote:So Kai might do some of the security topics, Wierdan some of the performance topics, VladSun other topics, etc.
Ah, I thought we are only talking about creating tutorials for Scaling/HA/Performance.