I agree totally. Its a function of graphics content by and large. An average screen is what? 2k of text (80x24 lines of characters), 1.4 megabytes of pixel information. (800x600x24 bit) Go figure!
Microsnot type web page generators also add HUGE amounts of (IMHO) needless rubbish in terms of styling and formnatting to persent even a few lines of text.
Used to be a problem - we used to up teh socket count on early Unix s=ystems to maybe 16,000 to stoip that happening. System V seems to have unlimited sockets though.
NT is of course crap, always was and always will be. More memotry leaks in that than holse in my jumpers..
I totally agree. We started off hosting a guy once, and by the time he has totally filled our 256k pipe - ad he had a LOT more than 10-15 concurrent users - he was making enough to take it up to Telehouse and host it all there.
Thanks that's given me some food for thought. I think I will set up some JMeter tests for the site I am currently working on as I would be interested to see the results. I keep meaning to put together some HttpUnit tests as well but you know what it's like. Manager (or SO in my current case) breathing down your neck asking you when the software will be ready. Sigh.
There is such a dearth of good (cheap) Java hosting I have considered setting up a company to provide it.
The collocation companies must be laughing all the way to the bank the amount they charge. Saying that though I looked at the price of getting a 1MB link run into our current house. I could buy a small island for the money they wanted.
I haven't had any down time (caused by my ISP - caused by my stupidity is a different matter) in the last year so you could say that I am pretty happy with the service.
Yeah, I don't remember the last time I used a HTML generator. I produce my pages using XML and XSLT (have a look here
formatting link
if you are interested to see my current project) which gives me the flexibility to produce content for any system. I'm looking at JSF at the minute but I don't think it is really suitable to commercial websites as you can't reliably bookmark pages.
I may be speaking out of the wrong orifice but as I understand it modern webservers like Apache can handle multiple clients per port hence the number of potential clients is no longer limited to the number of ports as it used to be (there was I believe a 1:1 mapping between ports and sockets hence the terms were used interchangably).
Now-a-days each new connection creates a socket that must be placed on an internal stack while the request is processed freeing up the port until the response is ready. As I understand, it load failures are now generally caused by an overflow of this stack.
Virtual Private Server
You share a box with others and have root access to your own virtual server.
Its actually straight forward with JMeter. It has a proxy recorder. You point your web browser at the proxy, and the proxy at the server, and just go and use your site. The proxy records all the actions you take along with all the parameters etc. So once you have done a set of actions you now have them stored as a test case. You can them have it replay that test multiple times with concurrency and random delays and offsets between them.
I found about half a dozen or more companies that did it, but after our bad experience with what looked like a very plausible service, I pulled a copy of the annual accounts for some of the potential replacement providers, and rejected any that looked like they were small companies (i.e. less than five staff) on the grounds that I did not want a repeat of the experience of not being able to get support when the do-do hits the fan big time! The end result was that if you want the professional
24/7 backup then you end up paying for it. The current provider, while expensive, is hard to fault in this respect.
I am in two minds now, they are expensive in relative terms, but given the staff and infrastructure costs they incur it is not too surprising (allow that they will be wanting to make a profit as well!)
I was not suggesting you will have a problem. Just more a case of if you are going to cop financial penalties should the site go down, you won't have the option of in turn claiming redress from the ISP if it was their fault.
I have also noticed that when I do lose ADSL connectivity, it is usually as a result of BT fiddling with something in the dead of night rather than anything the ISP has done. Normally it recovers all by itself, but on occasions a router reboot is needed.
You have just reminded me of the most ridiculous support call I have ever made. I had a cable Internet connection (with NTL - avoid them like the plague) in my previous place. One day it just upped and stopped working. I phoned support and quickly ascertained I knew more about the system than the support guy. He, of course, suggested I twiddle with some windows network settings (it was a simple Ethernet connection between the cable modem and the computer what could go wrong?).
I told him that I was running Linux so that would be hard. At that point he refused to talk to me till I told him I was running Windows. We went through his check list with me saying "Yes I've done that" at all the right moments. When we got to the end he must have had an attack of common sense and decided to talk to me. He was adamant that my computer was at fault but I had tried three different machines and all showed the same problem. The funny thing was NTL could ping the modem from their end and I could ping it from my end. Turns out the firmware in the modem had crashed. The guy asked how long the modem had been up so I told him about a year. He reckoned they didn't normally last more than a week between reboots and that I should have rebooted the modem at least once a day.
This puzzles me. PORT is used as a numbered endpoint. ALL traffic goes to the same PORT - port 80.
Sockets are not normally stack based. To create a socket is notmally a malloc() call to get memory from the memory pool. It is very rare for IP network code to do stack based allocation. I don't know of any thet dies.
Of couerse if memory is allocated in small socket sized blocks forever, ebentually fragmentaion, lack of real memory and so on will lead to thtasrhing as memory pages are swappe in and out of disk, and garbage collection takes over.
Early im0lenetaions of IP layers used fixed allocated memory blocks for sockets, and were probne to overload as a result.
The more modern web servers (zeus) don't handle concurrent connections on a single PORT and certainly not on a single SOCKET, what yhjey do do is multi-thread., so that only one copy of the progranm is needed for ALL connections, rather tha the old fashipned accept() and fork() type process to spawn a complete copy of the server for every request. That can and does lead to huge memory wastage, and possibly overflow of te concurrent process limits.
I do not know if the latest apache code is multi-thread.
Oh. I'd like to hear more about exactly what that actually means.
I stand corrected on a number of issues there - thanks for the info. Despite working with web servers for a number of years I must admit to not knowing an enormous amount about the exact inner workings (I will correct that).
What I had read was about the internals of Tomcat and the way it deals with multiple concurrent connections - I suspect I became confused at some point. As for the C stuff, well, us Java programmers wouldn't know a malloc from a fork in the head :-)
Apache 2.x has a number of workers (three common ones) one of which is fully multi-threaded and apparently very fast. Unfortunately a lot of people (perhaps even the majority) use PHP with Apache which has some non-thread safe libraries so the current recommended worker is the traditional non-threaded one.
Have you every used vmware or any other virtual machine emulator? Generally the way it works is, one, sometimes specialized, host operating system runs a number of virtual machines each in it's own memory space and with it's own virtual hardware.
With a VPS you have the whole virtual machine to play with which you can't tell from a real machine. The great thing about it is if you bork your virtual machine to the point where you can't get to it you can generally just login into your hosts website and (virtually) reboot or even reformat your machine.
IBM are into this sort of thing in a big way and have a customized version of Linux running on a 1024 node machine that can handle somewhere in the region of 10000 virtual machines which include Linux, BSD and Windows.
The IP stack keeps track of the different connections that it is running by means of sockets. A socket is uniquely identified by four separate bits of information: source IP, destination IP, source port and destination port. Typically the IP stack will allocate a "handle" to each socket to make dealing with it a little simpler. So each connection has exactly one socket. Multiple connections can be made to the same destination port.
The reason IP stack does not get confused when there are multiple incoming connections to a single port (i.e. the "well known" or "rendezvous" port such as 80 in the case of http), is because at least one of those four bits of information that make up a socket endpoint will be unique. Even when the same remote computer has initiated several connections (so the source and destination IP address, and the destination port will all be the same, the source port will be different in each case. This is because the calling stack will allocate the port number to the connection not the programmer). This is a so called "ephemeral port".
Modern software will typically create a number of "worker" threads to process each connection. Connections are stored in queue until a worker becomes available to process the connection. This allows the system to take advantage of multi threading to get maximum usage out of the interface bandwidth while at the same time not clogging the system by creating a thread for every endpoint.
Basically the way most hosting companies will provide your hosting unless you are prepared to stump up for your own dedicated co-lo server. One server hosts multiple sites all from the same IP address (this is why many of the internet content filters are so lame - they block the IP of a suspect site and take out fifty others that happen to be unlucky enough to share its server).
The server inspects the domain name that is a part of the requesting URL to work out which virtual site to invoke.
(this is why on smaller low traffic sites if you do a NSLOOKUP for the IP address of a site, and then stick the IP address into your browser, the web server barfs at you because it can't work out which site you are referring to).
That sounds more like virtual hosting rather than VPS. With a VPS you have a whole virtual machine to play with rather than just the resources granted to you by your ISP. You can install any software you want on it and do with it as you please. You can generally assign your virtual machine it's own unique IP address (it depends on the set up though) so reverse DNS will work.
My apologies, I was indeed describing virtual hosting, since I assumed (incorrectly) that was what you were referring to.
The IP address would have to be within a prescribed range that belongs to the ISP I presume... It would also sussgest that the server would then need to bind multiple IP addresses to its NIC so as to accept traffic for the VPSs.
I presume so. I haven't actually used a VPS system but I have used person virtual machines via vmware quite a bit and there is no problem with each one of those having its own IP address.
This is just using apache's innbult ability to accept ibncoming connections and allocate by name, rather than IP address to a bunch of different web server clones.
Full virtual macine implies you can load your own executable code on it and set your own permissions.
. Will put in a link to you. I keep meaning to build up a list of Derbyshire woodworkers so if you have any other links let me know. I use
formatting link
which gives cheap domain reg or transfer, web space, their own web page design tools (which I don't use), a control panel for easy upload of your site, or ordinary ftp access. All seems easy to use, cheap and perfectly OK. You don't need much space until you introduce complications like dynamic pages, other media etc.
Tomcat being a server side JSP (Java Server Pages) container and server. It can be used as a standalone server, however it typically is setup alongside Apache such that Apache processes all requests for static pages, but hand out .jsp pages to tomcat. It lets you host very sophisticated web sites and also supports a development model that largely lets you keep your web design / html elements distinct from your program code which makes maintenance much simpler.
I add my vote for them as well. I have just switched to them after using getdotted.com for many years. getdotted are ok, in so much as they haven't yet screwed up a domain, but they aren't the cheapest and they don't have a web interface to manage your domain.
If you need DNS services I strongly recommend everydns.net. They are free and very reliable. They are so good in fact I will donate this year to help keep the service going.
HomeOwnersHub website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.