Single or many WoW installs?

[Djarid]

hmm interesting... I think I will move to the 2 folder method

I had also already planned on using junction.exe but I get so little time at my PC atm that I would rather play than config

[Sarduci]

Disk seek times are normally the load time killer for games like this. Defragging and putting it on the out edge of the partition will help increase bandwidth, but 5 folders worth of wow is still more seeks than necessary.

That is why running one has to be more efficient than running 5 different installs. Seeking to read from 5 different locations at the same time vs reading data that's probably stashed in the cache from a single point on the drive should show a major decrease in overall seeks.

It is interesting to see all of the different setups that people do have.

[Chorizotarian]

hmm interesting... I think I will move to the 2 folder method

I had also already planned on using junction.exe but I get so little time at my PC atm that I would rather play than config

I tried this but had to give it up because of macros and addon settings from one clone stomping on the others. If you use symbolic linking you can get all of the benefits of running from a single directory without the pain.

[Nairi]

huge differene with 1 hd per wow sessions 8o

[keyclone]

if you are looking for performance, i'm not sure you'll see much of a gain by spreading files across the same drive. you still have the same bottleneck... a single read head on the drive

for performance, i would generally consider multiple drives and spread your wow folders across them.
this would give you multiple read heads focused on pulling the files, which would radically improve the read times.

i had posted a thread to start discussing such a solution here:
5x wow box spec (current: $1,639 +s/h+tax)

general wisdom seems to lean away from multiple drives today... with the SATA drives being enough.
i have not tried to compare yet... i'm still spec'ing. i figured you'd like to check out the info

oh, and welcome to the boards, Sarduci! nice first post

[BobGnarly]

I can't imagine you are going to see much if any difference in performance with multiple folders.

Keeping the wow files defragged and on the outer partition of your hard drive will definitely increase performance.

It's been mentioned already, but just to chime in, you will actually see a huge difference in performance - for the worse. I had a lot of problems with zoning and loading and whatnot until I moved everything to two folders (main account in one, slaves in the other). Much smoother since then.

[Mudd]

hmm interesting... I think I will move to the 2 folder method

I had also already planned on using junction.exe but I get so little time at my PC atm that I would rather play than config

I tried this but had to give it up because of macros and addon settings from one clone stomping on the others. If you use symbolic linking you can get all of the benefits of running from a single directory without the pain.

You need to setup junctions for the subfolders and not the entire wow folder. Basically you want to share the data files, and have seperate config files (wtf/addons). If I end up with some time I may try and setup a page on the wiki on how to do this. Be warned that you can accidentaly do weird things if you forget that folders are 'junctioned'. Patch day might cause you troubles if you launch 5 clients at once.

[Sarduci]

Well, here's the numbers from testing, averaged over 3 tests each of flying to the Wetlands then hearthstone back to SW. The data was taken from a 10 second snapshot after they all transported:

Single install: Max current disk queue length of roughly 43
Single disk using junction points: Max current disk queue length of roughly 43
Five installs, after defrags: Max current disk queue length of roughly 273

Average disk queue lengths we the same sort of result since sitting in the Wetlands gives me a disk queue length of 0:
Single install: Average disk queue length of roughly 6 at the end of 10 seconds
Single disk using junction points: Average disk queue length of roughly 6 at the end of 10 seconds
Five installs, after defrags: Average disk queue length of roughly 83 at the end of 10 seconds

For the non-techie folks, a current disk queue length of more than 8 per disk drive in your computer is unacceptable for high access applications such as databases. An average disk queue length of over 8 per disk over time is the same, not good for high performance. We're not exactly running a high performance application here, but anything above that baseline will tell you than you're system is waiting for your hard drives to catch up.

Now, for actual application of the numbers, it took less than 3 extra seconds for everything to show up and render on my system. Is three extra seconds going to matter for PVE, probably not. For PVP arena teams? Probably not. Normal in game performance worked just fine for all of them.

Read operations consisted of roughly 98% of the disk I/O that I tracked over the course of doing this all. Read and write disk I/O was all in short bursts, and considering how my page file was not being used for anything this is mostly (a little system and background things happening) straight WoW disk usage. The only time there were any kind of big or noticeable writes to disk were when shutting down clients. Addons will affect this number, the only things I currently have installed is x-perl. Running auctioneer or something that logs data like that may significantly change that percentage.

For all intensive purposes, single installs or junction points were the same. Multiple installs performed worse, but that is relative since the delay was minimal. You would probably find a bigger bottleneck on RAM or CPU on the average system. All in all, a wash in my book.

Space wise, single installs or junction points were the exact same size. Multiple installs were 400% larger than the single install or junction points. Personally I hate haven't redundant data on my systems.

Disk performance of a RAID1 for faster read rates would be better than RAID0; but the overall performance of a RAID1 measured against a single disk drive is the same as or worse in some areas such as write rates unless you are using a hardware based RAID controller that is running as store and forward vs a pass through configuration. All in all, I'd have to call that a wash.

Sorry it took longer than I thought to get this all, was a bit of a busy day on Sunday for work. Had to finish up the testing on the multiple installs on Monday night.

[keyclone]

@Sarduci

very interesting numbers. btw, where did you get the disk queue length?

did you try individual disks with an install on each? this would be different then stripped RAID as the type of access from 5 wows would all be targeting the same file, so disk contention would probably not work out well as they would all be queuing up for the same file (unless the caching helped.. which it should in theory... hmmm)

but as of now, a single install serving multiple wows on 1 drive is about the best performance you saw?

[Sarduci]

@Keyclone

No, unfortunately my last computer was all IDE, and I only purchased a single disk for this system. I was planning on adding more disks if I needed the for space more than anything since I'll be running a lot of virtual machines on there for work to look into 2008 server core services install. (Can't wait for the MS roadshow so I can get my free copy! [img]../forum/images/smilies/thumbsup.png[/img])

Disk queue length was taken from Perfmon in Windows since it's a built in tool to any install. It was logged to another system remotely so the disk I/O for permon would not show up in the charts. There were two physical disk counters that I used, current disk queue length and average disk queue length. Then I'd export it to a CSV file and use a VB macro I'd format the data and parse it that way.

Well, best performance was a mixed bag. I'd have to say that there is no clear advantage in single drive or RAID for my particular hardware. Less RAM or more instances running would change the numbers. Multiple installs were not as good in mine, but the overall performance was not bad.