Technical thread for Server Discussion (merged)

[Psionicist]

This thread must not die!

*Bump*

Michael Morris et al: Considered upgrading vBulletin? A couple of bugs have been fixed since 3.0.0 RC4.

 

[Krieg]

If you do not go with SCSI RAID you are making a mistake. Listen to what everyone is saying on disk seek times. This is critical.

The increased reliability is at least as important in my book.

 

[Psionicist]

The increased reliability is at least as important in my book.

Just to prove the point, here's another diagram (god I love these )

Access time (seek + latency + overhead) in ms. Lower = better.

 

[Kroax]

Psionicist, you need to learn to put some axis on your diagrams. Alright, acces time, but for what? A fruitbasket, a buss, an oven and your drawer?

I think I'm not the only one who would appreciate this.

 

[Krieg]

Look closely Kroax, they are there just hard to see as they are black on the grey background.

 

[Henry]

So Psionicist, you're saying for triple the cost, the +2 Millisecond seek time gain is worth it? For a work project recently, I recently priced an Ultra 320 SCSI drive (15K rot. speed) in the 73 GB range for $600.00 mail-order, that's pricey for 73GB, much less the 100 GB + range. A 300 GB 10K Ultra 320 SCSI was in the $1300.00 range.

Mind you, I'm not in charge of the tech purchase on the ENWorld server (We've got two other hardware professionals for that), but I'm just questioning your justification of speed versus cost, when SATA is pretty close. From all I've read, SATA may not outperform SCSI, but for the money it comes darned close.

 

[IronWolf]

So Psionicist, you're saying for triple the cost, the +2 Millisecond seek time gain is worth it? For a work project recently, I recently priced an Ultra 320 SCSI drive (15K rot. speed) in the 73 GB range for $600.00 mail-order, that's pricey for 73GB, much less the 100 GB + range. A 300 GB 10K Ultra 320 SCSI was in the $1300.00 range.

Seagate makes 15k SCSI drives with 3.6ms seek time. So that would be up to +4ms gain. Think of how many disk hits a database driven website may have in a given hour. Each ms gained from a fast seek on the HD is a ms second that can be spent doing something else.

Also, remember this isn't all about seek times. That is just *one* of the SCSI benefits. By going SCSI you also reduce load on the CPU processor by offloading some of the logic to the SCSI adapter. In addition with SCSI, the processor can request multiple blocks of data from the drive, go on to do other things while the SCSI adapter handles the queue.

From some of the orginal posts and what was in the news announcement yesterday, EnWorld is suffering from high CPU usage and wants to get a server that can last two to three years while sustaining growth of the website. SCSI drives can help with both, reduce processor usage by freeing the processor up for other requests while the SCSI controller takes on more work and allows for performance to remain high as the site grows over the next two to three years.

If you are building gaming machines, single high speed processors and SATA drive are great options. If you are building performance driven servers with two to three year lifespans, dual processors (or at least the option to add a second processor), a lot of quality RAM and a distributed array of high speed SCSI disks is the way to go.

Again, the EnWorld folks are doing a great job and it is much easier for me to comment from the outside looking in....

 

[Psionicist]

So Psionicist, you're saying for triple the cost, the +2 Millisecond seek time gain is worth it? For a work project recently, I recently priced an Ultra 320 SCSI drive (15K rot. speed) in the 73 GB range for $600.00 mail-order, that's pricey for 73GB, much less the 100 GB + range. A 300 GB 10K Ultra 320 SCSI was in the $1300.00 range.

Yes! SCSI is not just better access time and higher sustained transfer rates. SCSI-drives also have better multitasking and a more intelligent interface, SCSI can reorder requests for better performance. I think this speaks for itself: http://psionicist.online.fr/iometer_web.gif (again, that's the best SATA-drive right now)

A 36GB Seagate Cheetah (15000 RPM) is $275.00 at newegg. An Adaptec 29160 is $259.99, and that's a pretty expansive card compared to, say, Tekram. Total cost: 500 USD. That's 300 USD less than the gaming processor you have considered, and it will give you better performance.

Read the thread. I am not the only one who speaks in favour of SCSI here. Check out the dell servers. They are all dual cpus + SCSI. There is a reason.

And don't tell me it's overly expansive, because we just bought two DELL Dual Xeon 3.0 Ghz with SCSI-RAID (RAID 5) for less than $4500 USD total (that's 2250 each)

Edit: Sorry got carried away there, wrote that in problem solving/tech mode, and some people find my post offensive when I do that. No offense.

 

[Michael Morris]

This thread must not die!

*Bump*

Michael Morris et al: Considered upgrading vBulletin? A couple of bugs have been fixed since 3.0.0 RC4.

That is frontmost on my plate of things to do - though the functionality hacks we've added complicate matters. I've purchased a program called Beyond Compare to help with this matter.

 

[Psionicist]

We've also been donated an 8 month old Gateway server about halfway in power between our current box and what we intend to buy as a powerhouse, probably worth somewhere between $1000 and $1500. Talk about generous! As a result, we'll almost certainly be retiring our current machine and (after we purchase RAM and other accoutrements for it) using the donated Gateway as our box for handling the server software. An additional boost in speed will be the result.

So... What's the specs of the donated server, and what's the current "plan"?

Personally I would run Apache on the most powerful server, because MySQL is not the issue here (once you solve the disk-swapping issue with more RAM), PHP is. As I've demonstrated earlier in this thread, vBulletin does more PHP things than MySQL-things. So the best idea is to run Apache on the most powerful server.