Re: 2tier 250users max?

From: r.golds <r.golds_at_csdcsystems.com>
Date: Wed, 24 Nov 1999 07:16:17 -0500
Message-ID: <81gkqg$ra9$1@ssauraab-i-1.production.compuserve.com>

Generally speaking, I believe the idea is that if you can write stateless applications using some kind of web page serving transaction server then your chances of scaleablity by plugging in more hardware (and software) are good.

Over in the powersoft.public.easerver newsgroup they frequently say that it is difficult to deliver an application on MTS on a single hardware box with more than 200 concurrent users, so if you want 1000 users then you need 5 computers with MTS. I haven't got a clue why this might be. Nor do I have a clue what sort of box they are talking about or if it's some sort of limitation of the software. The reason I haven't got a clue is because I haven't deployed anything to the internet (well, one little Jaguar to Oracle thing this week but it doesn't count because I wrote it as a demo for ONE user at a time).

I note that sybase has a product 'EAS:Jaguar' which is a more-featured and more scaleable version of MTS. I haven't compaired the products myself, but IMO sybase has more experience delivering multiple large mutli user systems than Microsoft. Microsoft's focus is more on the smaller desktop and small business.

You could have a look at a comparison of products, and hence also derive things like # users per computer which you were asking about in articals like
"8 Web App Servers That Deliver. Products show good speed, scalability, but
tests conducted by Doculabs find differences in the number of users supported and fault tolerance"
http://www.zdnet.com/pcweek/stories/news/0,4153,409380,00.html where they say that they had 1500 users against four computers (each with four processors = 16 processors). The artical is mainly about pages pushed per second, although reading the chart it looks like the benchmark has each user reading 7 pages per second (3500 pages per second at 500 users). I don't read that fast myself.

As with everything else, numbers have to be taken with a grain of salt -- so here's some quotes from some of the loosers as defined in the chart in the article:

"Those numbers shocked everyone I know that has ever used MTS. MS was
allowed to bring in anything they wanted to and it is generally believed that the server/software used was seriously tweaked. If you look at the review in detail, it mentions some critical points that there was nothing in place for failover, etc. Basically, had anything gone wrong, the MS site would have fallen on its face."

"I was the engineer who wrote the Sybase solution for the PC Week benchmark.
Do seriously read the article. When you do you will find out

1) MS Used totally different web server hardware
2) MS Used totally different web server software
3) MS Used totally differnet database hardware
4) MS Used totally different database software
5) MS Used totally different app server hardware.

In essence that benchmark is comparing apples with rocks. You cannot look at the MS numbers along side anyone elses. I really wish PC Week had no put those two graphs on the same page. They dont at all deserve to be. I simply wish we had been allowed to compete with MS using the same hardware and database. But alas we didnt.
If you can take anything from that article it would be that Sybase was listed as one of the top performers amongst all app servers ever tested."

But of course, your question, and my answer, are not about who won or lost in some magazine-- but some approximates as to what you can expect. And since it will vary with your application, and the features and reliability you wish to support, the only real benchmarks will be the ones you do yourself.

Just my $0.02. There's probably a better way of finding out what you want to know.

Regards
Robb Received on Wed Nov 24 1999 - 06:16:17 CST