How do we move forward the right way? I'm thinking out loud to get input from people.

One possible option is to pursue some new systems for www.castlecops.com. A suggestion was made to use the Dell MD3000 (direct attached storage). At first glance it looks like a great notion, but a couple articles pointed out to me seem to cause concern:

http://blogs.smugmug.com/don/2007/10/01/dell-md3000-great-das-db-storage/
http://www.dslreports.com/shownews/Benchmarking-the-MD3000-powervault-under-linux-87401

The first one mentions MySQL, from their perspective, hasn't been tested on this DAS. One person suggested that MySQL and PostgreSQL are not capable of handling the DAS's dual write/read technology, and that Oracle is one which can. So MySQL might not be able to take full advantage. This alone makes one think, this DAS at least may not be worth pursuing. Would be great if it all worked.

Backplane bandwidth on the SAS is huge, so basically lots of bottlenecks would be eliminated.

Another possible is to go for two servers, probably dual cpu quad cores, or quad cpu quad cores with hardware raid, 16 or 32 GB RAM, an Areca card(s) instead of the 3ware. Try and eliminate the typical bottlenecks like bus speed.

And run either xfs under a linux install using suse or mandriva, or go with the zfs under freebsd 7. One box for mysql, the second for the web server, seeing that we rely heavily on libphp.

Hijacktrend hasn't launched because the hardware hasn't been able to support it. But a configuration like the above just might be what the application needs.

Right now we are at 2289 sql queries per second. And I'm sure folks are seeing some perf issues come up time and again.

So how do we do this the right way?

Speaking of MySQL, perhaps a year ago you were considering a migration to MS SQL. Have you dropped that consideration? Perhaps MS SQL would not have the same limitations that is attributed to MySQL.

There's also a discussion regarding which type of RAID array is best. Though the finding may only be applicable for the MD3000, it's leading me to wonder if a performance review of the current RAID type you're using would be useful.

I have no cycles for that kind of migration (db).

We're only mirroring right now.

Well, I am afraid I no longer have current experience in what you are looking at, but of course, I have an opinion!

1. Set a budget
2. Determine traffic load projections for the next 3 - 5 years
3. Determine HW/BW to meet those projections + another 100%.
4. Buy the equipment.

I think mirrored is the wise choice. You may take a slight performance hit with writes, but nice size drive buffers help with that. Reads, on the other hand, are very quick, as data can be pulled in from each drive in the array. And you have already reported that readers here read more than they type.

_________________
Bill, AFE7Ret
Freedom is NOT Free!

Right now they are mirrored, but going forward I don't know what is best.

This is also beyond any of my current experience. My gut feel is that with this level of RAM available, a Dual Quad, all in one box would be a good choice; RAM-RAM communication will always beat any backplane or network communications.

I don't have any feel re: drive configurations.

I agree with the SQL-Server/MySql: the migration would be a huge project in manhours. While SQL-Server is very capable, the licensing $$$ could be better applied towards hardware. And it *should* be possible to continue to tune this system to achieve very good performance under MySQL, although it may take repeated rounds of monitoring and tuning.

Your point on the extra RAM is a good one, and it has not been overlooked. The more RAM, the greater the chances of having the db run in memory without going to disk.

I Dunno if you happened to see my post about the DreamSpark offer from M$, but it does include SQL 2005: /postp1060752.html#1060752

I haven't fully read the license and it may not be usable for CC, but if you need to move from MySQL to MS SQL then you have access to at least one set of valid installations (from me)....

Also, reading up about SAS on Wikipaedia and about DAS on Wikipaedia makes me wonder - is the database extremely huge in size? If not, would perhaps SSD based on DRAM be an alternative for *just* the database(s)?

I realize that they are expensive as all get out - but if it came down to entirely new systems for a set budget, versus SSDs being attached to your current systems, I think the prices would be a lot more comparable....of course, I don't know jack about forums and databases and such so I may be way off base here, and if I am please excuse my ignorance (and take the time to enlighten me about a subject I know next to nothing about)....

I don't think cost would be a factor. I believe Paul has full license rights to MS SQL Server. I can apprecaite though that migrating to the MS SQL implementation would take lots of man-hours.

Some interesting memory storage technologies though.

I am not familiar with SSD, however, the current machine is only a single proc single core. We've reached some limit, in part due to the controller card as well. This main box will be upgraded to freebsd 7 and have zfs installed -- from what I understand.

if zfs, then perhaps raid-z is something to look into.

can you group/split the services this site provide, pirt mirt etc, from them forums. Have 2 site running in paralel on dedicated hardware with better hardware for services rather than forums. I don't if it is even possible

Cudni