Re: successful migration from client/server to 3 tier systems ?

In article <320F5F30.7E33_at_iwaynet.net>, btobin_at_iwaynet.net says...
>
>Casey Green wrote:
>>
>> # Microsoft has been posting some phenomenal numbers
>> # (like 5676 tpm-c under a simulated load of 5000 users) without
>> # the aid of a TP monitor. It seems to me that the claim that
>> # 2-tier doesn't scale is starting to look awfully thin.
>>
>> Now, I don't like to dispute the merits of Microsoft's
>> technology. So let's say that those numbers are bona fide.
>
> So far as I know, no one has suggested that cooking the TPC-C
 benchmark
>is even posssible. It's a very involved benchmark that simulates the
>operations of a bank branch.
>

The TPC does scrutinize the results, but do realize it's the final outcome that really is what shows. There are ways to "cook" how the results are achieved. That's not to say there is anything magical about what MS has done. There are plenty of other RDBMS vendors with 5000+ tpmC. And let's keep in mind that you can have 3-tier without a TP monitor. The Microsoft result used 7 pentiums as front ends. You might claim a custom TP monitor like program is just part of the application development and keep costs down by not including a TPM.

> Nowadays consultants and other
>assorted gurus are routinely specifying 3-tier architectures for
 projects
>that will never get within shouting distance of these numbers. The
 last four
>projects I've worked on have all been three tier, and all but one are
 never
>expected to be deployed to more than 150 users. Nevertheless, the
 most
>commonly cited reason for going three-tier is that "two-tier doesn't
 scale."
>Well, I think it's pretty clear from results like these that 2-tier
 scales
>pretty well. I wouldn't have any qualms about specifying a 2-tier
>architecture for, say, 1000 users-- a figure which I'd guess 80% of
 today's
>3-tier projects will never reach. The added costs of implementing a
 system
>as 3 tier are, in my experience, enormous: the available tools are not
 nearly
>as mature as the best of the currently available 2-tier products.
 That is
>rapidly changing, as everybody scrambles to jump onto the three-tier
>bandwagon, but it will continue to be true for the next year at least.

Don't get me wrong by the above comments Re: MS SQL Server tpmC results. You're discovering what I've known for some time, and was quoted in Open Computing July '95. TPM's are no panacea for performance problems. Some sites do get gains out of greater distribution of resources, but you do not need a TPM to do that.

You are correct in that many folks will not have the throughput to even test these principles. But when they do, be sure that C/S in general is a complex, interrelated, and not well understood environment. There are good products out there, and there are not so good ones. YMMV. If you choose the good ones, you're unlikely to get into too much trouble no matter what architectural ideosynchrasies you go for. But bad design can doom any project no matter what hardware/software you choose.

Just $0.02.

Geoff Crawford                              Phone:    (201) 627 - 0307
Innovative Client Servers                   FAX:      (201) 627 - 0634
24 Dogwood Drive                            Email: innov8cs_at_garden.net