MacroQuest64

Posted: **Tue Jun 21, 2005 11:23 am**

When you use Windows with a single hard drive and EQ, the game has to write windows information, eq .ini info, eq logging info AND your swap file all to the same drive.

If you add TWO hard drives to your system and move the swapfile to the second drive, EQ gets a significant perfomance boost.

If however you can afford it and buy a (small) THIRD drive, then windows info swapfile data AND eq .ini info can all be written at the same time, meaning you get lower zone times and better overall eq performance (especially during raids)

BTW the second and third drives don't have to be large..even a couple of older 10 or 20gb drives is plenty for a swapfile and/or EQ itself. (Just make sure your using ATA100 or ATA133 drives and cables for all your drives!)

Posted: **Tue Jun 21, 2005 11:34 am**

Or you could like...go nuts and turn on some caching, maybe..oh I dunno run a fairly modern multi-tasking OS with enough system resources and things configured such that this is a total non-issue.

psychosomatic != significant, HTH

Posted: **Tue Jun 21, 2005 11:40 am**

Your information is sadly incorrect. Or at least, not well described.

If you use two IDE hard drives on the same line, they're drawing from the same resources.

Additionally, as of several years ago, hard drives, operating systems, and motherboards have something called "cache". Cache uses memory to speed up hard drive reading and writing, which is thousands of times faster than your hard drive.

http://www.lavishsoft.com/wineq2/faq.php#2.1

2.1) Should I use the same folder for each session of my game, or separate folders?
Short answer:
Same folder

Long answer:
It is a common myth that game performance can be increased by using mulitple folders, possibly even on separate hard drives. The thinking goes that using separate copies of the files will allow them to be used independently and hopefully faster than trying to use the same files. However, this is not the case. A relatively simple concept in computer science says that hard drive accesses are slow, and memory accesses are fast -- about 10 milliseconds for hard drive access, and under 100 nanoseconds for memory access (as low as about 5ns depending on if the memory is stored on the CPU or if it is system RAM, etc), a ratio of 10 to 10,000,000. That's a pretty big difference. The point of explaining this is modern computer systems employ several "cache" systems to reduce the amount of file accesses as well as file access times by reusing unchanged data directly from memory, or by predicting the next data to be retrieved (for an easy to understand example, if you read the first half of a file, you can be expected to also read the second half). By using a separate folder, these cache systems cannot recognize that the data being read is actually the same. This means that instead of 10,000,010 nanoseconds for loading time for two sessions, it will probably be 20,000,000 nanoseconds because the entire data must be read twice. In terms of loading time, you have absolutely nothing to lose by using a single folder on a single hard drive, and can reasonably expect it to be faster than using multiple folders, regardless of whether it is on another physical hard drive. Note that it may be possible in some cases to see performance increases by using a separate hard drive, but the hard drives MUST be on a separate IDE channel, or they are competing for the same resources anyway (or use SATA, etc), and it is still unlikely that this will show improvement over a single folder.

Posted: **Tue Jun 21, 2005 3:59 pm**

WTF Lax, your version wasn't even surly

Posted: **Fri Nov 25, 2005 1:36 am**

i run dual 10,000 rpm sata raptors in raid 0... i have no problems with performance because of swap files..... i zone in a fairly quick amount of time

Posted: **Fri Nov 25, 2005 5:38 am**

iam_clint wrote:i run dual 10,000 rpm sata raptors in raid 0... i have no problems with performance because of swap files..... i zone in a fairly quick amount of time

RAID0 configurations are demonstrably slower with small file read/write times.
The sole time that that level of raid would increase performance is from large file read/writes.

The 10k rpm & speedy read/write on each drive compensate for the RAID0 performance degredation in your case.

The best, performance-wise, drive a desktop machine would use would be a 16Mb or higher cache, SATA, and supporting NCQ (along with a motherboard that supports this feature)

In other words, you wasted quite a bit of money on those nifty raptors.

Posted: **Fri Nov 25, 2005 4:40 pm**

I agree that spending cash on raptors for a gaming machine is probably a waste of money considering the cost vs capacity returns on 10,000 rpm drives, ie you get significantly more bang for buck in regards to drive size with 7200 rpm drives, I do not agree that raid 0 is actually a 'performace degredation'

More like .. probably wont notice a performance INCREASE but degradation is largely false.

There are many factors in determining optimum raid configuration and performance. These days, most drives come standard with an 8MB cache and some with 16MB caches, and unless you are filling that cache up on one single drive before moving on to the next drive to stripe the data, you probably arent seeing a performance increase.

stripe size = amount of data sent to a drive before moving on.
stripe width = the number of drives in the array

ideally where raid gets its performance increases is by overcoming the physical head to platter write times of the drive, but thats going to depend also on a couple of factors .. is it sequential or random reads / writes .. is it large amounts of data or small?

8mb of cache is a fairly significant amount of space for random writes to a drive, but I don't believe the data is written to the bus twice in any case so its hard to argue a performance degradation. It wasnt true of SCSI busses several years ago but I'll admit that I'm a few years out of the loop when it comes to SATA.

In any case typically what I used to find when I did things like raid testing for a living was that smaller stripe sizes (16K-32K) were overall yielded better performance when dealing with larger amounts of random reads and writes, such as what you would find in a RAID 5 array general file server or database server, but larger stripe sizes in a RAID 0 or RAID 4 array were more benficial when you had large amounts of large sequential reads and writes of large data files like in digital audio visual applications (typically 256K or greater stripe sizes)