!(*&%@# Rambus pisses me off

[superbaby]

Sorry for my typos, I didn't check over this one, I was in kinda a hurry.

For the Oracle DB, we run one like that too, except we are not limited to Java (although I love the JDBC interface, it's so easy to use). We use ColdFusion, Perl and several custom interfaces that we use to get/set tons of financial data. We have a connection pool of 100 active connections to the main DB and several of the redudant DB's have at least 50 connections.

We had a problem with the servers a couple weeks back in that we kept on running out of connections, too many queries were being sent to the server and the time it took to finish each one for was over 30 seconds (just the retrieval!). Because these queries were kept alive so long it forced new connections instead of being able to reuse old ones. So it was concluded that we had to improve the speed of the servers rather than purchase more connections (we hate giving Oracle money).

Several factors had to be evaluated, processor speed, HDD speed (big concern) and alternative memory types (i.e. Rambus). However it was concluded that the system was just too old and it would be a waste of time to upgrade them. Imagine 30 angry developers without their DB connections *shudder*.

So we brought in a benchmark analyst and asked him to give us a working spec of the fastest primary server he could configure. In the end he settled with a Quad Intel P3 Xenon solution, some SCSI stuff that cost over $15000 CDN and guess what... ECC SDRAM. He had built two servers, one with ECC SDRAM and one with RDRAM (basically he just swapped the motherboard, I don't know the details). The peformance gain over the ECC SDRAM config was about 2-3%, but the price difference was well over $6000!!!!!! (This is CDN BTW)

In the end we ended up with a new server that still used ECC SDRAM, simply because RDRAMs are not worth the cost at all. The reason we didn't use the Alpha again was that Compaq couldn't deliver within 2 months and we needed the server now for a major project.

So IMO I'm not speaking from gut feeling but from the results that the benchmark analyst gave us. It further affirmed my belief that RDRAM sucks ass and likewise their company.

I would run a couple benchmarks on those Alpha servers using RDRAM over ECC SDRAM, it might change your mind, especially when you see how much it costs!

 

[PCResources]

<< I would run a couple benchmarks on those Alpha servers using RDRAM over ECC SDRAM, it might change your mind, especially when you see how much it costs! >>



There is a HUGE difference between the implementation of RB in Intels design and the Alpha design, wich I have mentioned many times before, even in this thread, so just read my posts above, and you will understand the difference.

About benchmarking, i have already done that, the only bottleneck in this system is the memory. The memory they need will be more than 6000$ more expensive than an Intel/ECC SDRAM solution, but the Intel/ECC SDRAM combo you mentioned is not an option. They need the speed from the Alpha, and there is no other CPU/mobo offering that kind of memory bandwidth.

Thanks for your input though.

Patrick Palm

Am speaking for PC Resources

 

[ammon]

Yes Rambus originally filed in 1990. I have not seen any information about what was contained in that filing. In fact, some have said the original filing has been completely done over by ammendents over a span of 9 years! Apparently this is commonly accepted by the patents office.


This reminds of the the story about Alexander Graham Bell striking out his text and scratching in ammendments to his telephone patent after seeing a competitors design at the Philadelphia Centennial Exposition (use of graphite, I believe), which completely outperformed his design. His original design would not have worked.

The 1990 date doesn't prove they have not stolen IP.