“Honest question: What are the specific performance issues for a separate Web and Email machine?”
It’s really a number of connections in Open Transport (OT). If you have a website with eight artifacts per page, a person using Netscape will attempt all of those connections simultaneously and take up eight spots on the machine at once.
Email servers like Dan Knight* is using (Stalker Internet Mail Server, or SIMS) also attempt list distribution not one mail at a time, but by making multiple connections simultaneously out of the email software and constantly keeping one port open for incoming email. For this reason, Dan sometimes has trouble when too many connections are made at once, especially when SIMS tries to send out a couple dozen 1 MB attachments simultaneously to everyone on the list.
Since a Quadra 650 can handle around 50 to 60 simultaneous connections before it really overloads the OT stack, keeping the machine in the 30 to 40 range is optimum for a maximum. [It’s worse than that – I’m using a Mac IIfx. dk] WebStar allows limitations on the number of incoming connections, as does SIMS. But this doesn’t solve the problem.
When a connection to WebStar is open, the host machine (web server) loads the artifact into memory, then transfers it to the network through the port. Let’s say that a web page has eight high quality artifacts – all 300 KB pictures, say for a realty organization selling houses on the Web, and they’re using higher quality images in transfer. At the moment that the browser hits this page, it gets the primary HTML and connects to the web server eight times simultaneously and pulls down these pictures. The web server immediately loads eight 300 KB images into memory, moving memory usage to an extra 2.4 MB of RAM.
If multiple people are accessing this site, the connections could jump to 40 simultaneous connections on this page – meaning perhaps 12 MB of memory is being used by the web software at this very moment, either in application space or system heap (usually system heap). Now it doesn’t always happen this way, simply because WebStar and other web software may realize it’s sending the same information to multiple people, but as a web server gains more and more diverse web pages, the odds that caching will assist in this become really low. And poor web designers like to use blank spacers in their tables and borders with hog connections and increase the memory usage on a web server.
Combine this with a web software package that has no idea that the email package will be gouging up memory the same way, and it should be rather obvious that dedicating separate machines for web and email might be a way to go – if not now, then probably in the future when memory and connection problems begin occurring. You could always go with a combined solution – usually from StarNine, who makes several integrated packages and packages that know about each other running to solve these problems, but in the case where SIMS would be using heavy load, the web server portion might not be given the priority it needs to serve properly. And dedicating a specific minimum of threads to the web server software would be inefficient when the email server is under heavier load (4 a.m.) and the web server is not. See the problem?
- Scott L. Barber <firstname.lastname@example.org>
- Pres/CEO, SERKER Worldwide, Inc.
- Providing Hardware/Networking/Telecomm for 13 years
Scott L. Barber posted this to Quadlist, the listserv for users of 68040-based Macs, on September 10, 1998. It is reprinted with his permission.
* Dan Knight, listmom of Quadlist and publisher of Low End Mac. At the time this was published, we were handling email for several lists and serving up web pages on a Mac IIfx. As noted, email was handled by SIMS. Web pages were served by NetPresenz. That’s all the IIfx does. I later set up a dedicated email server for Low End Mac to handle all of our Mac-related email lists. dk
Keywords: #webserver #emailserver #mailserver #sims
Short link: https://goo.gl/a3pwKb