I am choosing a better web server for big SPA application with dozens of JS and css files. With HTTP/2 we are now able to not merge them into two big files (3 MB for JS), that take pretty long time to load when on slow connection. But which server is better for the new HTTP/2 paradigm?
Nginx was designed to solve http/1 problems, and it's advantage was better serving numerous connections, with HTTP/2 there is only one connection for all the files, so the feature seems redundant now. What do you think, what can you advise me?
That's a very subjective question, and probably not a good fit for StackOverflow so imagine this will get closed. But here's my two cents...
Full disclosure: I primarily use Apache.
For a start let's address one of your incorrect points: Nginx wasn't designed to solve HTTP/1 problems. Nginx was designed to solve some of the scalability problems of previous web servers by being based on an asynchronous, event-driven model. Under HTTP/2 there should be less connections per client, which you could argue makes scalability less of an issue as each client uses only 1/6th of the resources they did previously - but that's probably a little simplistic. Apache has an event-driven MPM module for years now too (though often not turned on by default in case of any thread unsafe PHP applications - but this would also be a problem with Nginx!). This brings them more back in line, though there's still a lot of debate about this and many say Nginx is still faster. In my experience, unless you are dealing with truly huge volumes (in which case you should be looking at CDNs, load-balancers and cache accelerators), few will ever notice the difference between Nginx and Apache. This is especially true when downstream applications and systems come into play - a slow PHP application will quickly negate any performance or scalability issues at the web server level.
Anyway, back to your main question:
For HTTP/2 support, my choice would be Apache over Nginx. It has had better HTTP/2 support for some time. Nginx only added HTTP/2 Push support in early 2018 for example, whereas Apache has had that for a number of years now. Apache also supports a PushDiary (based on the now-abandon Cache-Digests proposal) to prevent pushing resources that have already been sent, supports 103 Early Hints for pushing early, and push prioritisation options. Moving on from HTTP/2 push, Apache also supports using HTTP/2 in proxy mode (though it's still marked as experimental and the usefulness of this is questionable at the moment), and HTTP/2 over HTTP (h2c - though again usefulness is questionable since browsers do not support this). I also find the main developer of the Apache HTTP/2 implementation very responsive on the GitHub page for the mod_http2 module (included as part of core Apache since 2.4.18 and no longer marked as "experimental" since 2.4.26).
On the flip side, I understand that Cloudflare uses a customised Nginx based web server, and they have HTTP/2 push for over a year now (it was them that backported this implementation to Nginx). So, given Cloudflare's scale, that speaks volumes to the implementation of that though not sure how customised it is from the core Nginx code.
There is also a HTTP/2 conformance Testing tool available and when I ran this against some common HTTP/2 servers (for a book I wrote on the subject btw) I got the following results which clearly shows Apache as the most compliant with the spec:
Now to be fair, most of the causes of errors are in not responding correctly to bad requests, that in a perfect world should never be sent anyway so aren’t that important. But still, we don’t live in a perfect world, and error checking is an important part of technology so I for one would certainly prefer the more compliant server. Similarly as pointed out in the comments below, the tool and web servers themselves, can be subject to race conditions and other problems which may incorrectly report errors.
Ultimately you are best choosing the implementation you are most comfortable with. The general feel has always been that Nginx is lighter and easier to configure, but on the flip side perhaps isn't as fully featured as Apache because of that. HTTP/2 support seems to continue that theme. If you want to play with upcoming HTTP/2 features then, to me, Apache definitely has the edge at the moment (though nothing to say that won't change in the future). However, for the basic use cases of HTTP/2, they probably can be considered similar. Even HTTP/2 Push is not used much yet, and there are serious concerns it could degrade performance if not used appropriately or due to implementation issues, which is probably why it has not been a priority for Nginx and while they only implemented it later.
Related
There are several discussions around the web about some "anonymous" benchmarks released by Litespeed Technologies on http2benchmark.org (etc) allegedly showing that their LS servers are faster than Nginx LEMP stacks (apparently, they knowingly disabled FastCGI caching):
https://hostedtalk.net/t/litespeed-servers-publishing-fake-benchmarks/4431
https://www.reddit.com/r/selfhosted/comments/f06vse/litespeed_servers_seem_like_a_marketing_scam_are/
https://community.centminmod.com/threads/is-lsapi-litespeed-faster-than-php-fpm-nginx-etc-or-not.19153/
My question is three-fold:
As far as LSAPI on Litespeed vs PHP-FPM on Nginx is concerned, what would be a "fair" way to compare performance re: processing PHP queries since the environments are different?
Besides the PHP handler, is there any other part of the Litespeed stack that can be accurately benchmarked at the application-level (e.g. LS Cache) instead of the stack?
Have any unbiased third parties compared these two stacks in an accurate fashion?
I'd rather post this as a comment but it's pretty full already up there.
This benchmark routine by Ike Devolder is probably one of the most fair and detailed that I've come across in the past few years comparing Litespeed and Nginx:
https://blog.herecura.eu/blog/2020-06-16-openlitespeed-vs-apache-vs-nginx/
Some of the confusion is re: Litespeed vs. OpenLitespeed, but for this purpose it doesn't really matter since the biggest difference with Nginx is LSAPI, the PHP processor, rather than using PHP-FPM, and since OLS supports the same server-level caching that Litespeed does...
One of the best things about Ike's review is that he doesn't definitively say one is "better" or "faster" and instead talks about different components of either stack, while monitoring resources.
The TLDR for 2021 is probably something like this:
Both Nginx and Litespeed serve static assets extremely well
Litespeed as a stack scales higher than Nginx currently, partially due to HTTP/3 support... but 99% of websites probably will not notice this since we are talking about insane traffic levels, which almost never reach the origin server these days due to things like CDNs or remote MySQL servers, and since HTTP/3 is not fully integrated into the internet yet... in fact, even in Ike's own tests he seems to have tested both internally (not over i.e. UDP) which further reinforces this point since HTTP/3 is UDP-only. However, regardless of protocol, Litespeed does appear to scale to a higher level than Nginx... keep in mind, this is not the same as saying that LS delivers requests "faster" than Nginx, because real world requests at a typical traffic level are probably indistinguishable in terms of speed/response times. It does mean that i.e. a shared hosting environment with 500 cPanel websites would probably scale better using Litespeed... so it makes sense why that seems to be Litespeed's main target.
Litespeed appears to use more RAM memory than Nginx in general
Nginx appears to use a lot more CPU than Litespeed at high traffic levels
TCP (not Unix socks, the Litespeed default) performed a lot better, and this is true for Nginx as well, since it avoids disk I/O... so, using default Litespeed settings is still a bad idea
Anyway, best not to get too detailed, since software changes all the time really. The next version of Nginx will probably support HTTP/3, among other features. I do wish that the Litespeed guys would let the unique quality of their software speak for itself, and chill out on the aggressive (inaccurate) claims being made... but most of their web hosting partners probably benefit from that hype, which is maybe why not many people are questioning the claims being made...
I have a medium sized website called algebra.com. As of today, it is ranked 900th website in US in Quantcast ratings.
At the peak of its usage, during weekday evenings, it serves over 120-150 queries for objects per second. Almost all objects, INCLUDING IMAGES, are dynamically generated.
It has 7.5 million page views per month.
It is server by Apache2 on Ubuntu and is supplemented by Perlbal reverse proxy, which helps reduce the number of apache slots/child processes in use.
I spent an inordinate amount of time working on performance for HTTP and the result is a fairly well functioning website.
Now that the times call for transition to HTTPS (fully justified here, as I have logons and registered users), I want to make sure that I do not end up with a disaster.
I am afraid, however, that I may end up with a performance nightmare, as HTTPS sessions last longer and I am not sure whether a reverse proxy can help as much as it did with HTTP.
Secondly, I want to make sure that I will have enough CPU capacity to handle HTTPS traffic.
Again, this is not a small website with a few hits per second, we are talking 100+ hits per second.
Additionally, I run multiple sites on one server.
For example, can I have a reverse proxy, that supports several virtual domains on one IP (SNI), and translates HTTPS traffic into HTTP, so that I do not have to encrypt twice (once by apache for the proxy, and once by the proxy for the client browser)?
What is the "best practices approach" to have multiple websites, some large, served by a mix of HTTP and HTTPS?
Maybe I can continue running perlbal on port 80, and run nginx on port 443? Can nginx be configured as a reverse proxy for multiple HTTPS sites?
You really need to load test this, and no one can give a definitive answer other than that.
I would offer the following pieces of advice though:
First up Stack overflow is really for programming questions. This question probably belongs on the sister site www.serverfault.com.
Https processing is, IMHO, not an issue for modern hardware unless you are encrypting large volumes of traffic (e.g. video streaming). Especially with proper caching and other performance tuning that I presume you've already done from what you say in your question. However not dealt with a site of your traffic so it could become an issue there.
There will be a small hit to clients as the negotiate the https session on initial connection. This is in the order of a few hundred milliseconds, will only happen on initial connection for each session, is unlikely to be noticed by most people, but it is there.
There are several things you can do to optimise https including choosing fast ciphers, implementing session resumption (two methods for this - and this can get complicated on load balanced sites). Ssllabs runs an excellent https tester to check your set up, Mozilla has some great documentation and advice, or you could check out my own blog post on this.
As to whether you terminate https at your end point (proxy/load balanced) that's very much up to you. Yes there will be a performance hit if you re-encrypt to https again to connect to your actual server. Most proxy servers also allow you to just pass through the https traffic to your main server so you only decrypt once but then you lose the original IP address from your webserver logs which can be useful. It also depends on if you access your web server directly at all? For example at my company we don't go through the load balanced for internal traffic so we do enable https on the web server as well and make the LoadBalancer re-encrypt to connect to that so we can view the site over https.
Other things to be aware of:
You could see an SEO hit during migration. Make sure you redirect all traffic, tell Google Search Console your preferred site (http or https), update your sitemap and all links (or make them relative).
You need to be aware of insecure content issues. All resources (e.g. css, javascript and images) need to be served over https or you will get browsers warnings and refuse to use those resources. HSTS can help with links on your own domain for those browsers that support HSTS, and CSP can also help (either to report on them or to automatically upgrade them - for browsers that support upgrade insecure requests).
Moving to https-only does take a bit of effort but it's once off and after that it makes your site so much easier to manage than trying to maintain two versions of same site. The web is moving to https more and more - and if you have (or are planning to have) logged in areas then you have no choice as you should 100% not use http for this. Google gives a slight ranking boost to https sites (though it's apparently quite small so shouldn't be your main reason to move), and have even talked about actively showing http sites as insecure. Better to be ahead of the curve IMHO and make the move now.
Hope that's useful.
For a SaaS running on Node.JS, is a web-server necessary?
If yes, which one and why?
What would be the disadvantages of using just node? It's role is to just handle the CRUD requests and serve JSON back for client to parse the date (like Gmail).
"is a web-server necessary"?
Technically, no. Practically, yes a separate web server is typically used and for good reason.
In this talk by Ryan Dahl in May 2010, at 37'30" he states that he recommends running node.js behind a reverse proxy or web server for "security reasons". To elaborate on that, hardened web servers like nginx or apache have had their TCP stacks evolve for a long time in terms of stability and security. Node.js is not at that same level yet. Thus, since putting node.js behind nginx is easy, doesn't have many negative consequences, and in theory increases the security of your deployment somewhat, it is a good choice. At some point in time, node.js may be deemed officially "ready for live direct Internet connections" but wait for Ryan/Joyent to make some announcement to that effect.
Secondly, binding to sub-1024 ports (like 80 and 443) requires the process to be root. nginx and others automatically handle binding as root and then dropping privileges to a safer user account (www-data or nobody typically). Although node.js has system call wrappers in the process module to drop root privileges with setgid and setuid, AFAIK other than coding this yourself the node community hasn't yet seen a convention emerge for doing this. More on this topic in this discussion.
Thirdly, web servers are good at virtual hosting and in general there are convenient things you can do (URL rewriting and such) that require custom coding in node.js to achieve otherwise.
Fourthly, nginx is great at serving static files. Better than node.js (at least by a little as of right now). Again as time goes forward this point may become less and less relevant, but in my mind a traditional static file web server and a web application server still have distinct roles and purposes.
"If yes, which one and why"?
nginx. Because it has great performance and is simpler to configure than apache.
I currently have one server with nginx that reverse_proxy to apache (same server) for processing php requests. I'm wondering if I drop apache so I'd run nginx/fastcgi to php if I'd see any sort of performance increases. I'm assuming I would since Apache's pretty bloated up, but at the same time I'm not sure how reliable fastcgi/php is especially in high traffic situations.
My sites gets around 200,000 unique visitors a month, with around 6,000,000 page crawls from the search engines monthly. This number is steadily increasing so I'm looking at perfomrance options.
My site is very optimized code wise and there isn't any caching (don't want that either), each page has a max of 2 sql queries without any joins on other tables, indexes are perfect as well.
In a year or so I'll be rewriting everything to use ClearSilver for the templates, and then probably use python or else c++ for extreme performance.
I suppose I'm more or less looking for any advice from anyone who is familiar with nginx/fastcgi and if willing to provide some benchmarks. My sites are one server with 1 quad core xeon, 8gb ram, 150gb velociraptor drive.
nginx will definitely work faster than Apache. I can't tell about fastcgi since I never used it with nginx but this solution seems to make more sense on several servers (one for static contents and one for fastcgi/PHP).
If you are really targeting performance -and even consider C/C++- then you should give a try to G-WAN, an all-in-one server which provides (very fast) C scripts.
Not only G-WAN has a ridiculously small memory footprint (120 KB) but it scales like nothing else. There's work ahead of you if you migrate from PHP, but you can start with the performance-critical tasks and migrate progressively.
We have made the jump and cannot consider to go back to Apache!
Here is a chart showing the respective performances of nginx, apache and g-wan:
g-wan.com/imgs/gwan-lighttpd-nginx-cherokee.png
apache does not seem to lead the pack (and that's a -Quad XEON # 3GHz).
Here is an independent benchmark for g-wan vs nginx, varnish and others http://nbonvin.wordpress.com/2011/03/14/apache-vs-nginx-vs-varnish-vs-gwan/
g-wan handles much more requests per second with much less CPU time.
NGINX is the best choice as a webserver now a days.
The main difference between Apache and NGINX lies in their design
architecture. Apache uses a process-driven approach and creates a
new thread for each request. Whereas NGINX uses an event-driven
architecture to handle multiple requests within one thread.
As far as Static content is concerned, Nginx overpasses Apache.
Both are great at processing Dynamic content.
Apache runs on all operating systems such as UNIX, Linux or BSD and
has full support for Microsoft Windows & NGINX also runs on several
modern Unix-like systems and has support for Windows, but its
performance on Windows is not as stable as that on UNIX platforms.
Apache allows additional configuration on a per-directory basis via
.htaccess files. Where Nginx doesn’t allow additional configuration.
Request Interpretation-Apache pass file System location. Nginx
Passes URI to interpret requests.
Apache have 60 official dynamically loadable modules that can be
turned On/Off.Nginx have 3rd Party core modules (not dynamically
loadable).NGINX provides all of the core features of a web server,
without sacrificing the lightweight and high-performance qualities
that have made it successful.
Apache Supports customization of web server through dynamic modules.
Nginx is not flexible enough to support dynamic modules and loading.
Apache makes sure that all the website that runs on its server are
safe from any harm and hackers. Apache offers configuration tips for
DDoS attack handling, as well as the mod_evasive module for
responding to HTTP DoS, DDoS, or brute force attacks.
When Choose Apache over NGINX?
When needs .htaccess files, you can override system-wide settings on
a per-directory basis.
In a shared hosting environment, Apache works better because of its
.htaccess configuration.
In case of functionality limitations – use Apache
When Choose NGINX over Apache?
Fast Static Content Processing
Great for High Traffic Websites
When Use Both of them -Together
User can use Nginx in front of Apache as a server proxy.
Apache has been the de facto standard web server for over a decade, but recent years have brought us web servers that consume less RAM and handle many more requests per second using fewer threads and asynchronous i/o. In my opinion, I also find the configuration of these servers to be more straightforward and minimal.
Why do people use Apache when asynchronous servers are so much more lightweight? Is there any clear benefit?
Ubiquity, "good enough", and familiarity.
Apache's .htaccess provides flexible configuration. This allows users on a shared host to customize certain settings of an apache without having to alter the core apache configs.
It is the standard server bundled in typical LAMP setups, although, many services use other web servers for in conjunction (like static files, video streaming, etc.).
Since Apache is popular, it's easy to find a solution to any problems.
Other than that, other solutions would probably be better.
Apache IS asynchronous if you want it to be with the Event MPM. Unlike Nginx and Cherokee, etc., it is not the default.
Apache's made some important moves in 2.4 so it can be more competitive — esp. as it pertains to serving static requests using the Event MPM. Various benchmarks don't speak well of this, but:
It's very difficult to ascertain how much slower Apache is in 2.4 because Apache's out of the 'box' configuration is detrimental to performance and legacy holds it back in some respects. For example, .htaccess requires stat/reading a multitude of files on every request, which may include many rules and regexes. Nginx doesn't have this problem, nor does Cherokee. Litespeed has .htaccess support in Apache's style, but only if you pay for it. Most benchmarks don't turn off features like those.
Most of the benchmarks are also ridiculous in that they're run locally and over a GbE network or similar. A real web server has to cope well with various speeds, including 3G phones. It could be that Apache's performance is better in the real world.
I doubt it.
Nginx is still faster, and I might choose it, but Apache isn't asleep.