New cracking potential coming soon....

[cautery]

Couldn't resist a deal I found on HDDs and the incredibly low RAM prices.... Besides, I have this beautiful brand new GENIE server case just sitting in the closet doing nothing, so...

I am in the middle of building a new MSI 694D Pro AR based dual system in the GENIE w/ 4 IBM 40GB drives configured at ATA/100 RAID 0+1. The OS will actually reside on an IBM 10GB drive on the ATA/66 controller.

Maybe NOW, I can really stretch the legs on the two 700's I have in my current dually.

I'll be resurrecting the matched pair of C300a's I had in the Tyan Tiger before I got the PIII-700s. You know, those WarpCore parts just keep making there way back to chew more blocks!

It'll take me a while to get this new sys up as I am stretched pretty thin between work, and several projects.

Later...

 

[HomerSapien]

Sweet.. Good things never die.

 

[Kilowatt]

Hey Clay,
Welcome back.
For those of us that want to know, are you going to be cracking under the old "WarpCore" banner, or starting anew?

On a side note, my WarpCore node is still alive and well, the board is running a pair of PIII450's @600 being used everyday as my Main office W2K Pro workhorse, and the processors are in a BP6 at home running my Linux box there.

Good to see ya around again.

Kilowatt

 

[efun]

call be young but what is WarpCore??? TIA

 

[cautery]

Kilowatt! Good to hear from you dude! Glad to see that your node is still working/cracking for you (whatever number of systems it has spread to).

efun - Back in early '99, a bunch of us Team AnandTech folks banded together in The Collective, under the name, The WarpCore Project. Together, we built WarpCore which was a Cluster of Workstations (COW, get it? ). It consisted of a total of 24 dual processor diskless nodes in a custom made rack remote booting from a LINUX server. All of this hardware was dedicated to cracking RC5 for Team AnandTech. It was tons of work, but worth every moment, and more fun than I've had in a long time! There is a RealMedia file out there called "engage.rm" that was made of the live launch.

I like to think that TWP and The Collective are partly responsible for Team AnandTech's success in getting to the "top of the heap" in the RC5 project....

Though I'm not real happy with the current state of affairs!

Someone want to explain to me why/how those Dutch wannabe AnandTechers (friendly gig, OK. ) have bested us on the stats list? <grrrr> Are there really that many people in the Netherlands?

Later...

 

[ZapZilla]

Awsome cautery!

I believe I have one of the old WarpCore mobo's and it's still active, though its crunching OGR now.

Go Team AnandTech!

 

[cautery]

Hey ZapZilla! How's it going?

Update: Got my coolers in the post today... Now if the board, NIC, and a few other odds and ends would arrive!

I spent last evening putting all the hard drives in the case. Looks pretty cool with all those IBM drives stacked up in there.

Also finished the power supply mods for the most part. Desoldered EVERYTHIN from the board, shrink wrapped the ATX harness, and added back two independent sets of shrink wrapped power wires to power the front and back of the case. Upped the wire gauge to 16GA to handle the larger current draw on the harnesses... and of course to avoid excessive voltage drop at the end devices. Just left them bare for now, as I am going to custom fit the connectors to the exact config of the case when everything is finally installed.

Later...

 

[Windogg]

4 x IBM 40GB HDD and 1 x 10GB? Why not use Windows 2000 Server's software RAID5 abilities? I snagged 3 x Maxtor 40GB HDDs and scraped together 2 x 3.5GB Quantums and did just that. The overhead will be lower. You would only lose 25% of total capacity as opposed to 50% with 0+1. I tested it and it worked perfectly. I lost on the the drives early (FedEx did a heck of a number on the package). The drives operated 4 anxious days in &quot;Redundancy Failed&quot; mode while the dead drive was being RMAed. When the new drived arrived, I popped it in and the array rebuilt itself in a few hours. Not too shabby at all. CPU utilization during writes was around 10 - 30% and read hovered around 1 - 10%. Performance is excellent with multiple video stream going across the network without a skip. PM me for more details.

Windogg

 

[cautery]

PM en route Windogg, though I am unable to see how a software solution could beat a hardware solution.

 

[Fardringle]

Um, because it's free, whereas a good RAID5 card will cost more than an entire crackrack system (or two or three)!

Besides the fact that I've never even heard of an IDE RAID5 card, and since they already have the IDE drives...


If your work/livelihood depended on it, then hardware would definitely be the way to do, but if you're just doing it to say that you did (or for non-essential use) then software is usually sufficent.

 

[cautery]

Fardringle, who said anything about a RAID 5 card? I think we're getting a bit far afield here.

I am using an MSI 694D Pro AR motherboard that has a Promise ATA/100 RAID controller on it capable of RAID 0, 1 and 0+1. I have chosen to do 0+1 with 4 x 40GB IBM drives and use a 5th IBM drive (10GB) as the OS/admin drive off the ATA/66 IDE controller.

My point in my previous post assumed too much (that everyone had read the configuration posted back a ways) was that I don't see how a software RAID 5 solution could beat a hardware 0+1 solution. Frankly, I don't need a parity drive. Nothing I do is that critical. Mirrored drives is way more than I have EVER had. Unless I'm realy missing something...

My main reason for using 0+1 is faster reads and writes.... that's it.

Later...

 

[Windogg]

Yes IDE RAID5 card exist but they cost just as much their SCSI counterparts. Check out the Adaptec AAA-UDMA. The reason they are expensive is because almost all hardware RAID (IDE/SCSI) is based on the Intel i960 RISC processor. The cards are pretty neat since there is usally a DIMM or SIMM slot for extra memory. Most IDE &quot;Hardware&quot; RAID card just add extra ports with firmware that used the CPU for all the data crunching. Notice how there is no memory slot for cache. That is because your system is doing the work.

Windogg

 

[Windogg]

Screenshot of RAID5 in Windows 2000 Server

 

[cautery]

I understand now Windogg.... never thought about it that way. I am seriously considering getting an eval ATA RAID 3, 5 card from Promise to test out. It has up to eight chnnels and is a TRUE hardware card as it has both the i960 onboard processor and cache memory slot(s).

It's not QUITE as expensive as a SCSI RAID card and the IDE drives are much less expensive.

For now, I'm going to test whether Win RAID 5 software or the Promise ATA RAID controller meets my needs best as I said in the PM.

Opening this question up publicly:

What tools should be used to test the array performance? Preferably free tools... I'd rather NOT have to spend a chunk of change on HDD tools if I don't have to... but I'd like the tools to have some legitimacy though...

Thanks...

 

[jatwell]

Clay, are you looking for fault tolerance or just speed?

If you use the 0+1, you'll have an 80GB, fault tolerant array (Two mirrored 80GB striped arrays). You can also get that capacity and fault tolerance with RAID 5 with only three drives. You also said &quot;Frankly, I don't need a parity drive. Nothing I do is that critical.&quot; so I'm not sure if you need fault tolerance. If not, doing the mirroring would just waste 80GB

Now you could just have a 160GB RAID 0 non-fault tolerant partition, which is what I would probably do if you don't have mission critical data you're storing. Not to mention it would be pretty darn fast

 

[cautery]

jatwell - I probably didn't express myself accurately and had some misconceptions about the Promise Controller. Here ia how I see it:

My priority is better speed in BOTH reads and writes. My secondary is fault tolerance, which should probably be my first priority considering how long I go sometimes between backups. The data on this array, while generally replaceable, would be a huge pain in the but to regenerate.

Initially, I had settled on the 0+1 setup which would give me and 80GB data array mirrored on the other 80 GB array. According to Promise and since all drives are identical, I should get about twice the write speed of a single drive, and approximately 4 times the read speed, thouhg the sustained read speed will be ATA/100 bus limited I suspect.

If I went all out for the RAID 0 setup, I'd get a 160GB array with zero fault tolerance (actually a higher likelihood of data loss due to any of four drives failing causes loss of whole array). I'd still get smokin' reads, and the writes would be 3 times the single drive speed. I don't need 160GB right now, and I don't think I can accept INCREASED risk over a single drive of data loss.

Finally, per Windogg's suggestion, I can install Win 2000 SERVER rather than Pro (since I own both) and build a software RAID 5 using all four drives, giving me a 120GB fault tolerant array (via parity drive... for some reason I thought RAID 5 was mirrored AND a parity drive... guess I was mistaken). The array should have pretty good read speeds, but takes a hit in the write department.

What I have decided to do is set Server up (apply service packs, updates, patches, etc... yadda, yadda), make a Partition Magic backup or the OS drive, then create the data array under Win RAID 5. Test the array with as yet to be determined testing tools. Then blow the array away and create an array using the Promise controller. Test. Then make a decision on which method to use.... if the write hit on RAID 5 is small enough to overlook, etc, etc.

Thanks,

 

[jatwell]

Clay, looks like you definitely have a handle on it

Have a look at this article. I've tried RAID before and find this article to pretty much tell the real story behind RAID. A LOT of good info there and it dispells some common misconceptions about RAID and also about ATA33/66/100 (Just read that article linked in the RAID story)

http://www.realworldtech.com/page.cfm?ArticleID=RWT022201021241