Many of us have web and application servers that use plain TCP.
Some of us have web and other servers that use a secure layer such as SSL.
My understanding of SSL is that the handshaking is very computationally intensive, and the encryption of an ongoing connection is (relatively) cheap.
My assumption for you to correct: an average hosting box (and info on what is average at cloud hosting would be cool too) might easy be expected to be able to saturate its network connections with AES-encrypted packets, but have difficulty doing a thousand RSA handshakes per second. Client authentication with certificates is substantially more expensive for the server than anonymous clients too.
What kind of rules of thumb for the number of session setups per second for SSL are there?
Why not just measure? It will give you real numbers on the exact software and hardware that you are using. You'll also be able to measure the impact of changes in the server infrastructure (adding more boxes, SSL accelerators, tweaking parameters, what have you).
You are correct that you would be hard pressed to get to a thousand SSL handshakes per second on a single box. In fact, I'd say it's probably impossible. A few dozen per second, not a problem. A thousand, not without a lot of $$$.
It's also likely that you don't really need 1000 handshakes per second. That's quite a lot, and you'd already need quite a lot of traffic to need something like that: See this: What do I need in SSL TPS Performance?
Remember that normally you won't be doing new SSL handshakes all the time. Browsers do the handshake once, and keep the connection open over a number of requests and/or page views, so your needs for handshakes per second may be much lower than you think.
As Ville said there is no real option then to try it out on your configuration. But don't underestimated the symmetric encryption of data after establishing a link. It might be less expensive but if you are going to download a lot of data over the encrypted channel than it might cost a lot more than the initial negotiation.
So for this you have to build a common scenario for the usage of your site and then stress test.
Related
The architecture I have is that clients receive events from a service through websockets. We are building another service which also needs to communicate it's events to the clients. We are faced with a design decision - whether to change existing service or just add another websocket connection from client to new service.
It is technically complex and expensive to change the other service as it's not maintained by us, but at the end of the day, if necessary, it is possbile. Adding another connection to new service would be cheap and no redesign would be needed, but it would imply that there would be two connections in parallel.
Other questions i've looked at often get answers from server perspective, where TCP connections are limited and thus it's important to reduce the amount of connections, but it's obviously not applicable here, since it's two separate services. What i'm asking is that how big of a difference it is from a client perspective that instead of a single websocket connection there would be multiple connections? My view on this has been that our goal should be to use the same connection as each connection reduces performance, but to be honest it's just an opinion and not a proven fact.
Since it's enterprise setting, backwards compatibility and performance is highly important, but at the same time, development cost is way higher if we'd choose to go with one connection instead of two. So.. is there a real difference or it's something I should not worry about?
From a client perspective, on any modern platform, the difference in cost of managing one vs two sockets in an application or browser should be negligible. It sounds to me like simplifying development and maintenance effort should dominate your decision.
This article mentions and this site seems designed to show that HTTPS can be faster than HTTP. I'm surprised; I thought HTTPS was just HTTP plus encryption, which adds a small, likely negligable amount of work but doesn't remove any.
Why might an HTTPS page load be faster than one over HTTP?
It's a bit of a con to be honest.
HTTPS is slower than HTTP. There's no denying that. HTTPS works over HTTP so has to do everything HTTP does and more. Now, with good web server config, the computational cost of HTTPS is almost non-existent to the average user on today's modern hardware but it is there. But it also slows down the first page render as it takes a few hundred extra milliseconds to set up the HTTPS connection. Again not a big deal for most people but it is there.
Now there is the argument that someone - be it a mobile network or ISP or whatever - can change HTTP by injecting ads and the like, potentially slowing down a website, but that's not the reason for the speed difference here.
The reason that website is faster is because it is using HTTP/2 when using HTTPS and not when using HTTP. HTTP/2 is faster than HTTP/1.1 - especially for websites with lots of resources.
Of course you can say that HTTP/2 is only available over HTTPS and while that is true*, the corollary is not - implementing HTTPS does not automatically give you HTTP/2.
*Well technically it's not true that HTTP/2 requires HTTPS as per the spec, but all the browser makers have said they will only support this over HTTPS so it basically is true to all intents and purposes.
Additionally the sample website loads 360 small and near identical (but crucially not identical) resources. Precisely the sort of thing that HTTP/2 is very good at. And while average web pages are growing, most of them don't load 360 near identical images - so that network latency is basically the only bottleneck. Most have other issues as well that are nothing to do with the network latency issues that HTTP/2 massively improves.
The speed gains for HTTP/2 are hugely impressive and it is the future and everyone should use it, as latency is a major bottleneck. But that test site is an extreme example of it. Depending on the exact site's make up, HTTP/2 will mostly offset the cost of HTTPS and in many cases more than offset it - but that does not mean HTTPS itself is faster.
There are very good reasons to use HTTPS, and the article is fantastic for listing them all (except for that first one). In my opinion HTTPS should be the default and everyone should move to it - precisely for the other reasons listed. But it's a lie to say HTTPS is faster that HTTP. Or, at the very least, it obfuscates the truth by not explaining why it can be faster. And then listing HTTP/2 as a second, seemingly unrelated, reason to further confuse the reader! I just don't understand why they couldn't combine these two points into one and fully explain this so questions like this didn't need to be asked? Same for that sample site - why is there no FAQ to explain why HTTPS is apparently faster?
Historically yes, https was http+ssl/tls, so it was slower
But now with spdy/http2, it's a new protocol, which can be faster than http when dealing with multiple requests:
it can compress headers, and if you send the same header multiple times (like cookies) it doesn't need to send it fully but just an id
if can reuse tcp connections, so it avoid the overload of opening multiple tcp connections and stream efficiently data
If you use some kind of network scanner (e.g. a component of an antivirus, proxy of firewall), it may scan plain HTTP traffic causing a slowdown. At the same time, it won't touch encrypted HTTPS traffic unless you installed a special root certificate that will help the intermediate scanner to process HTTPS traffic. So if there is some kind of intermediate service scanning HTTP traffic, but not HTTPS -- using HTTPS will be much faster.
I remember years ago, one of the reasons for not using SSL was the it used a lot of resources, so it affected the performance of applications.
Nowadays, with the current technologies, is this still a point to bear in mind?
This question arose as a workmate is concerned that using SSL will hinder the performance of his application.
Why? The idea is that there will be thousands of clients that will be opening temporary connections every some determined time frame (I think it's set to 1 minute). So he's concerned that all the authentication process of all those clients is going to be very power consuming and affect the performance of his application. The other alternative is to use permanent connection so the authentication is done only once, but the CTO still hasn't decided which method we'll be using (last notice was temporary, hence this question).
The question is ill-formed. If you need security, you have little choice but to use SSL, and so comparing it to plaintext is completely pointless. If on the other hand you don't need security, you don't need SSL.
However I did an extensive experiment over the Internet some years ago, which showed that SSL was roughly 3x as slow as plaintext.
In the last 4 years I have seen (benchmarked) iPhone AES encryption speeds increase 13x on iPhones. Speeds are also dependent on the data length since there are two parts: the setup and data encryption/decryption.
As usual benchmark your usage and judge if there is a performance issue.
As #EJP states, if you need security you need to use https (TLS) encryption.
I was considering what would it take (technologically) to move all the web traffic to HTTPS. I thought that computers are getting faster, and faster, so some time from now it will be possible to run all traffic via HTTPS without any noticeable cost.
But then again, I thought, encryption strength will have to evolve to counter the loss of security. If computers get 10x faster, encryption will have to be 10x stronger, or it will be 10x easier to break.
So, will we ever be able to encrypt all web traffic "for free"?
Edit: I'm asking only about the logic of performance increases in computing vs encryption. If we can use the same crypto algorhytms and keys in 20 years, they will consume a far lower percentage of the overall computing capacity of a server (or client), and in effect, that will make it "free" to encrypt and sign everything that we transmit over networks.
One of the big issues with using HTTPS is that its considered secure and so most web browsers don't do any caching, or at least do very limited caching.
Without the cache, you'll notice that HTTPS pages load significantly slower and a non-encrypted page would.
HTTPS should be used to protect sensitive information.
I have no idea about the CPU impact of running everything through SSL. I would say that on the client side, the CPU isn't an issue since most workstations are running idle most of the time anyway. The big program would be on the web server side due to the sheer number of concurrent requests that are being handled.
In order to get to the point that SSL is basically 'free', you'd have to have dedicated hardware for encryption (which already exists today).
EDIT: Based on the comments, the question's author suggests this is the answer he was looking for :
Using crypto is already pretty fast,
particularly considering that we're
using CPU cycles vs. data
transmission. Crypto keys do not need
to get longer. I don't think there's
any technical reason why this is
impractical.
-David Thornley
UPDATE: I just read that Google's SPDY protocol (designed to replace HTTP) looks like it will use SSL on every connection. So, it looks like Google thinks that it's possible!
To make SSL the underlying transport
protocol, for better security and
compatibility with existing network
infrastructure. Although SSL does
introduce a latency penalty, we
believe that the long-term future of
the web depends on a secure network
connection. In addition, the use of
SSL is necessary to ensure that
communication across existing proxies
is not broken.
Chris Thompson mentions browser caching, but that's easily fixable in the browser. What isn't fixable on switching everything to HTTPS is proxy caching. Because HTTPS is encrypted end-to-end, transparent HTTP proxies don't work. There are a lot of places where transparent proxying can speed things up (for instance at NAT boundaries).
Dealing with the additional bandwidth from losing transparent proxying is probably doable - allegedly HTTP traffic is trivial compared with p2p anyway, so it's not as if transparent proxies are the only thing keeping the internet online. It will hit latency irrevocably, and make a slashdotting even worse than it is currently. But then with cloud hosting, both those might be dealt with by tech. Of course "secure server" takes on a different meaning with cloud hosting, or even with other forms of de-centralisation of content across the network like akamai.
I don't think the CPU overhead is that significant. Sure, if your server is currently CPU bound at least some of the time, then switching all traffic from HTTP to HTTPS will kill it stone dead. Some servers may decide that HTTPS is not worth the monetary cost of a CPU that can handle the load, and they will prevent literally everyone adopting it. But I doubt it will be a major barrier for long. For instance, Google has crossed it already and happily serves apps (although not searches) as https without fuss. And the more work servers are doing per connection, the less proportional extra work is required to SSL-secure that connection. SSL can be and is hardware accelerated where necessary.
There's also the management/economic problem that HTTPS relies on trusted CAs, and trusted CAs cost money. There are other ways to design a PKI than the one SSL actually uses, but there are reasons SSL works how it does. For example SSH places the responsibility on the user to obtain a key fingerprint from the server by a secure side-channel, and this is the result: some users don't think that level of inconvenience is justified by its security purpose. If users don't want security, then they won't get it unless it's impossible for them to avoid it.
If users just auto-click "accept" for untrusted SSL certificates, then you pretty much might as well not have it, since these days a man-in-the-middle attack is not significantly more difficult than plain eavesdropping. So, again, there's a significant block of servers which just aren't interesting in paying for (working) HTTPS.
Encryption would not have to get 10x stronger in the sense that you would not need to use 10x more bits. The difficulty of brute force cracking increases exponentially with an increasing key length. At most key lengths would have to get slightly longer.
What would be the point of running all traffic through SSL, even stuff where there is obviously no advantage? This seems incredibly wasteful. For example, it seems ridiculous to download a Linux distro through SSL.
The cost isn't that great nowadays.
Also...having a computer that is 10x faster will in no way make it necessary to change encryption. AES (a common encryption for SSL) is strong enough that it would take a very very long time to break.
Will it be possible? YES
Will it be advisable? NO
For a few reasons.
extra cpu cycles on server and client would use more power which incurs cost and emissions
ssl certs would be required for every server
it's useless to encrypt data that doesn't need to be hidden
IMO, the answer is no. The main reason for this is that if you consider how many pages have items from multiple sources that would each have to use https and have a valid certificate that I don't think would work for some of the big companies that would have to change all their links.
It isn't a bad idea and maybe some Web x.0 would have more secure communications by default, but I don't think http will be that protocol.
Just to give a couple of examples, though I am from Canada which may affect how these sites render:
www.msn.com :
atdmt.com
s-msn.com
live.com
www.cnn.com :
revsci.net
cnn.net
turner.com
dl-rms.com
Those were listed through "NoScript" which notes this page has code from "google-analytics.com" and "quantserve.com" besides the stackoverflow.com for a third example of this.
A major difference with https is that a session is kept open until you close it. Saves a lot of hassle with session cookies but puts a load on the server.
How long should google keep the https session with you alive after you send a query?
Do you want a persistent connection to every popup ad?
Our company runs a website which currently supports only http traffic.
We plan to support https traffic too as some of the customers who link to our pages want us to support https traffic.
Our website gets moderate amount of traffic, but is expected to increase over time.
So my question is this:
Is it a good idea to make our website https only?(redirect all http traffic to https)
Will this bring down the websites performance?
Has anyone done any sort of measurement?
PS: I am a developer who also doubles up as a apache admin.
Yes, it will impact performance, but it's usually not too bad compared to the running all the DB queries that go into the typical dymanically generated page.
Of course the real answer is: don't guess, benchmark it. Try it both ways and see the difference. You can use tools like siege and ab to simulate traffic.
Also, I think you may have more luck with this question over at http://www.serverfault.com/
I wouldn't worry about the load on the server; unless you are serving high volumes of static content, the encryption itself won't create much of a burden, in my experience.
However, using SSL dramatically slows down web sites by creating a lot more latency in connection setup.
An encrypted session requires about* three times as much time to set up as an unencrypted one, and the exact time depends on the latency.
Even on low latency connections, it is noticeable to the end user, but on higher latency (e.g. different continents, especially Australasia where latency to America/Europe is quite high) it makes a dramatic difference and will severely impact the user experience.
There are things you can do to mitigate it, such as ensuring that keep-alives are on (But don't turn them on without understanding exactly what the impact is), minimising the number of requests and maximising the use of browser cache.
Using HTTPS also affects browser behaviour in some cases. Certain optimisations tend to get turned off for security reasons, and some web browsers don't store objects loaded over HTTPS in the disc cache, which means they'll need to get them again in a later session, further impacting the user experience.
* An estimate based on some informal measurement
Is it a good idea to make our website
https only?(redirect all http traffic
to https) Will this bring down the
websites performance?
I'm not sure if you really mean all HTTP traffic or just page traffic. A lot of sites unnecessarily encrypt images, javascript and a bunch of other content that doesn't need to be hidden. This kind of content comprises most of the data transferred in a request so
if you do find feel that HTTPs is taking too much out of the system you can recommend the programmers separate content that needs to be secured from the content that does not.
Most webservers, unless severely underpowered, do not even use a fraction of the CPU power for serving up content. Most production servers I've seen are under 10%, even when using some SSL traffic. I think it would be best to see where your current CPU usage is at, and then do some of your own benchmarking to see how much extra CPU usage is used by an SSL request. I would guess it isn't that much.
No, it is not good idea to make any website as only https. Page loading speed might be little slower, because your server has to perform redirection operation unnecessarily for each web page request. It is better idea to make only pages as https that may contain secure/personal/sensitive information of users or organization. Even if the user information passing through web pages, you can use https. The web page which have information that can be shown to all in the world can normally use http. Finally, it is up to your requirement. If all pages contain secure information, you may make the website as https only.