PDA

View Full Version : Comments, RAID Data Base



ricks
04-20-2002, 11:54 PM
Please post your comments, thoughts and concerns on the RAID Data Base here.

TZ
04-21-2002, 08:01 AM
Add controller; single channel or dual (put a check in one or the other column); also bring the SoftRAID / Apple RAID / ExpressStripe into its own column; OS (10.1, OS 9)?

There is a definite need. I tried to find info in the ftp archive and was frustrating. I look in the xlr8yourmac and storagereview databases and this would help fill in a gap. I thought the Reliability database, how many drives, how long in use, when/if it ever failed.

Because RAID needs to use 'identical' drives, have the drive info as ST336752LW (http://www.seagate.com:80/support/disc/specs/scsi/st336752lw.html) and then separate column for number of drives, total RAID size?, drive capacity?. I would really like to see the drive model revision have its own column and not just "2G X15" but the Seagate model number. The way www.storagereview.com (http://www.storagereview.com) will present a table and let you click on a drive model and have it bring up the drive specifics and review they did. That is ever so helpful - and the way people search.

Gregory

PS: I managed to 'break' the 53MB/s ceiling on B&W using just two X15's, one on each channel of 39160.

[This message has been edited by Gregory (edited 21 April 2002).]

Doc
04-21-2002, 11:10 AM
Would it be helpful if we also considered a comparison against hardware RAID boxes?. For instance, I have one of those Arena IDE RAID boxes coming in this week that I'll be bolting onto a Miles2 card on my ASIP server. I could compare its performance against my 10K Cheetahs.. (the Arena will have 6 x 100 GB 7200 RPM drives in it).

Chris

gmidd
04-21-2002, 11:21 AM
Separate columns maybe for driver vendor & version and for testing utility used for benchmarks?

TZ
04-21-2002, 11:39 AM
A column for HDST benchmarks. I like theirs and wish a) it was free, b) that the ATTO be "phased out."

Also, Apple Disk Basher will be improved and there needs to be an OS X set of tools, and a way of showing only OS X; comparing OS X to OS 9.

Does a user enter the benchmark name? or select on a list of benchmark utilities (other?).

It was the relational database that made the mainframe accessible. The demand on resources grew exponentially though. So the idea was to off-load to departmental 'minis' PC's and workstations. Then the "PC" took over.

All from the creation of a simple spreadsheet.

Gregory

[This message has been edited by Gregory (edited 21 April 2002).]

ricks
04-21-2002, 01:29 PM
I like it, keep it comin. Somewhere is the right combination of details to mix into the spread sheet.

I have issues at first glance with having several test programs doing the benchmarking. Each bench tests using different methods and comparing them is comparing apples and oranges. Then graphing those results is a waste of time.

I also am not sure how to handle recommending a for purchase product versus ATTO's and MacBench freebies. I have never used HDST, this is the first I've ever even seen it mentioned as a reliable testing solution.

These are my concerns and this is why I didn't just write a set of guidelines for sending in your data, I still don't know the answers. You guys are the only way I'll solve this, the total experience on these forums is awe inspiring!

I see the need to 'keep it simple stupid' this problem. The fancier the data base gets the less it will be legible. At the time we started this, and really still today, there is no acceptable bench program for testing arrays and drive performance in OSX. I see a need to cure that. So we did all our testing with ATTO set to 8mb samples size and MacBench at default settings in OS9.2.

I didn't add the MacBench scores to the spread sheet yet because the data didn't fit the graphs that were the main reason I put in the spread sheet in the first place. Using the spread sheet for a guide was an afterthought from Kaye, I was just using it for graphing when he set up the SS version you see.

I am open to almost any suggestion that comes out answering questions at a glance versus cluttering and confusing things.

I also totally agree that a separate column needs to state host card model/firmware. Maybe a separate driver column. Hard drive details. I can see no way to keep it on one spread sheet if we use multiple test programs, those would have to be kept separate. If we start testing in OSX, actually WHEN we start testing in OSX those results will also need to be in a separate SS.

Doc, I see great need to have commercial arrays posted on the SS. It's wonderful that you're just now getting it since testing is so mush easier before you put it on line. I'd like to have performance numbers from many many arrays here. THe need is just as real for hardware RAID numbers.

Thanks again for the input. With a little head banging and tossing ideas around we'll get a decent resource out of this. Keep it up.

Rick

Doc
04-23-2002, 04:52 PM
Raid box has arrived - basically it's the ones that Maxtronic sells via their resellers..

It came with 6 x 100 GB WD drives that I had to install - the problem I immediately ran into was that the drives weren't recognized by the machine (sort of). They were there on boot but you couldn't configure the RAID. Chatted with the reseller who was relatively oblivious and who told me to contact "whoever has their phone number on the inside cover of the manual" - nice...

Anyways - I asked other local gurus who have the same hardware RAID boxes and they immediately said - if you have WD drives, pull the jumpers off" Done deal and all was fine..Somehow, the jumper, which is set up initially so that the drive defaults to cable select, must screw up the backplane... No idea if same thing happens with other makes. Oh - we also had to download and update via a VT100 terminal connection, a firmware upgrade since we were using these big IDE drives.

Anyways - works now - and it's currently beeping as it initializes the RAID.

Nice looking box though.. I'll hook it up to my Miles2 card shortly (once it finishes beeping) and give a test...

Chris

Doc
04-23-2002, 04:54 PM
Oh - part two - did you guys want me to test this out on an OS X box?. I have a hankering to give it a shot but I'll need advice on which SCSI card to put into it (this RAID has the LVD option)

Let me know

Chris

crazyeights
04-24-2002, 11:10 AM
Rick said:

"I am open to almost any suggestion that comes out answering questions at a glance versus cluttering and confusing things. "

KISS is good Rick, I wholeheartedly agree. One of the principal benefits of having this tool is to make it user accessable (sp?). If the database needs as much or more Moderator feedback to the end user, in order to have the data make sense, it sorta defeats part of the purpose, doesn't it?

Elegant simplicity makes for the best tools. I don't know squat about RAID, but I was able to look at that table and "see" the bottom line.

I'm not clear on what Gregory meant about the mainframe to PC comment. But when I was thinking about a way to kill 2 birds with 5 stones, the relational database model of organizing the data makes some sense, as long as it didn't take on a life of it's own.

What a fantastic idea ...........you and Kaye are a couple of righteous dudes Rick!!!

my meager and uneducated 2

http://macgurus.com/infopop/emoticons/icon_biggrin.gif



[This message has been edited by crazyeights (edited 24 April 2002).]

Doc
04-24-2002, 11:13 AM
more feedback on this RAID

Points to keep in mind

GRANITE cables are the best - the no-name crap stuff that shipped with this thing were useless. It took me a few tries to narrow it down but now I'm good to go and have it up and running.

Ditto for the terminator - GRANITE rules...

Now - if someone can explain why my write transfer rates are so terrible ... I'm getting about 60 on my reads (under the ATTO test w/ 8 MB sample sizes))...but awful write rates..

Chris

Doc
04-24-2002, 11:56 AM
Here are some numbers for this RAID box

RAID 5 configuration - Western Digital UDMA 100 GB / 7200 RPM drives - 6 drives set up as a nominally 500 GB array

Formatted with the Atto driver from Express Tools - Formatted size: 463 GB (thereabouts)

Tested under OS 9.1 / PTP 225 upgraded with an XLR8 G3/500 / Atto Express tools

Hooked up with an Initio Miles2 w/ version 1.05 of SmartSCSI

8 MB sample size

Peak Read: 58.39 MB / s; Sustained READ: 58.11 MB / s
Peak Write: 27.04 MB / s ; Sustained Write: 16.49 MB / s

512 KB sample size

Peak Read: 66 MB / s; Sustained Read: 56 MB / s
Peak Write: 33 MB / s ; Sustained Write: 34 MB / s

I chose the simple default settings so it's not set up for PrePress or Digital Video etc...

Compare that to my 18 GB 10K LVD (Cheetah) drive on the same bus (SoftRAID driver - no RAID set up).

Peak Read: 33.09 MB / s; Sustained READ: 31.58 MB / s
Peak Write: 32.19 MB / s ; Sustained Write: 31.59 MB / s

The write performance is terrible - any ideas on what I can try out? There's nothing on this RAID now so I could swap it into whatever config you'd like to have tested (0, 1, 3, 5)

Chris

ricks
04-24-2002, 12:13 PM
Doc,

Post all the particulars like computer model, OS, scsi card, a link to the specs page on the RAID box maybe..

Maybe something will stand out as the problem. Kaye and Gregory both have dug up parameter changes that can make huge differences in RAID performance. Not sure how much controllability your RAID5 software gives you.

You should expect a large slowdown during writes. That's when the controller is creating parity data and writing to disk more data than you read back. Might be some way to optimize the settings on the segments.

I got your email, the 8 MB setting is appropriate for the sample size in the test.

I have been collecting the results from running Apple Developer Tools 'DiskBasher' utility. We might also give that a try and do some comparisons in OSX. Interesting utility in that I haven't really figured out the right combination of configurations that will tell me what I want to know about an array. Almost too many options if you can believe that. I need to figure out how to post where to download the test utility from, hard to link to the ADC since it's a logon site, need to check with Magician before I post it on the ftp server.

Rick

Doc
04-24-2002, 12:57 PM
Here's a link to the actual manufacturer page: http://www.maxtronic.com/

and the one I got:
http://www.maxtronic.com/new_html/small_bussiness_raid.htm

It's the industrial EX II as a rack-mount

Chris

Doc
04-24-2002, 12:59 PM
Oh - btw RAID level is set in hardware - there's an IDE -> SCSI backplane on the box that allows you to use IDE drives but have it appear to the computer as an LVD drive...


I can get at some parameters by "dialing" in via a serial cable and pretending to be a VT100...

Chris

TZ
04-24-2002, 03:03 PM
Disk Basher is on ftp site and is a "work in progress" but it does represent an attempt to provide a more "real world" benchmark, though right now it is overly optomistic I hear. Find Apple Performance Tools Suite (ftp://ftp.apple.com/developer/Tool_Chest/Testing_-_Debugging/Performance_tools/) at ftp.apple.com/developer/tool and online manual Using Performance Tools (http://developer.apple.com/techpubs/macosx/Essentials/Performance/PerformanceTools/Using_the_P_mance_Tools.html)

Gregory

[This message has been edited by Gregory (edited 24 April 2002).]

Doc
04-24-2002, 05:41 PM
On a suggestion from Dan at Initio, I'm going to try out something later - I'm going to reset the RAID to a level 0 setting and test out my R/W transfer rates.. see if they get better - the fact that it's creating that parity thing during the write stage is what's hurting the performance... Raid 0 should give me the best outcome.

I'll keep everyone posted..

Chris.

Doc
04-25-2002, 07:51 AM
G'morning all

After resetting the hardware RAID to a level 0, I just tested it using ExpressTools

System: Power Tower Pro 225 running an XLR8 G3/500 , OS 9.1, Miles2 card
RAID: 6 x 100 GB WD 7200 RPM IDE drives running as a single 563 GB drive using a Maxtronic Industrial II EX hardware RAID (appears as single SCSI drive to the rest of the world connected via LVD)

Test conditions 8 MB transfer size

Peak read: 58.71 MB / s; Sustained read: 58.49 MB / s
Peak write: 33.85 MB / s ; Sustained write: 24.84 MB / s

I should have realised this earlier but thanks to Dan and all who suggested that the RAID setting was the culprit.

Next step - test it out on a QS under an Atto card...

Chris

kaye
04-25-2002, 02:20 PM
I'll make some changes to the SS data page and shoot it to Rick for posting and more critique. k

ricks
04-25-2002, 03:17 PM
Doc,

Thanks for all the great testing. Your posts really help show how and what the setup issues are with that Hardware box. BTW, if you all will send me that there raid box I'd be happy to test it on QS and UL3D for a bit. http://www.macgurus.com/ubb/tongue.gif

Rick
Ever the 6 year old.

kaye
04-26-2002, 02:21 PM
We have a revised SS. I started with Gregory's comments but see that gmidd has raised some ideas that I will change. I am trying to keep the screenshot as small in width as possible so that those with smaller monitors can still see it in its entirety.

Since the Intech QuickBench has a cost attached, for now anyway, we are going to stick with ATTO EPT, MB5, and DiskBasher. Open to more thoughts though. Keep in mind that this whole screenshot idea was just to get things going in the interim while a different method is being worked on.

Keep the suggestions coming. And if I missed a suggestion, repeat it. k

Doc
04-26-2002, 02:30 PM
Looks great....

I can run some benchmarks on both the hardware one as well as several simple Miles2 LVD arrays on my 7300 systems.

Chris

kaye
04-26-2002, 02:43 PM
I knew I forgot something. For now, testing with DiskBasher, we would like to see:

Disk Cache: 0 bytes, Number of iterations: 100, Sequential, File multiple: 1.

We want the 8388768 (8MB) stats. So for example, PR = MAX (MB/s), SR = AVG (MB/s), PW = MAX (MB/s), SW = AVG (MB/s).

So for example:

Results assume 800.0 MHz processor(s), 132.9 MHz bus, OS10.1.3
Volume: 120GXP HardRAID, Disk Cache: 0 bytes
Carbon Version: 1.4.0, Number of iterations: 100

HFS+, Read, No Cache Pref, Sequential, File multiple: 1
SIZE (bytes) MAX (MB/s) MIN (MB/s) AVG (MB/s) SIZE (KB/MB)
8388768 112.16 79.01 110.71 8MB

So PR=112.16, SR=110.71.

HFS+, Write, No Cache Pref, Sequential, File multiple: 1
SIZE (bytes) MAX (MB/s) MIN (MB/s) AVG (MB/s) SIZE (KB/MB)
8388768 65.34 63.59 64.74 8MB

So PW=65.34, SW=64.74

k

TZ
05-08-2002, 07:07 AM
How would comments be included? I'm running a 39160 in the normal video 66 mhz slot on my b&w right now.

Gregory

kaye
05-08-2002, 10:05 AM
Good point, Gregory. I'll leave that to Rick but I suppose in the Host Card column. The entries are capable of wrapping to more lines so I suppose somthing like this would result:

Host Card
Adaptec39160
66MHz video slot

or
Adaptec39160
66MHzVideoSlot

Anyway, something like that. Rick? k

ricks
05-08-2002, 10:35 AM
Another morning. Not sure I'm awake again...

The column widths are auto scaling, if the host card needs more information the column will get wider as well as wrapping the text. I'm not sure how to handle the UBB code having priority over window widths yet, haven't tested many configurations.

I will find out what happens if the card line get's significantly bigger on a couple of entries. Some information, such as your running in the Graphics slot, is essential.

Can always rewrite a little of the code...........

Rick

TZ
05-10-2002, 04:32 PM
Rick,

did you ever put the UL3D in the 66 mhz slot in your Yosemite? ATTO didn't think putting the UL3S in there would help, and that they have one card that has "-33" on the serial number, another that is "-66" and designed ONLY for 66 mhz slots. I tried but didn't really see any benefit.

Just got around to setting up 39160 in 66 mhz slot. Two 2G X15 on one channel. Finally, sustained writes of 115MB/s. No PCI-to-PCI bridge bottleneck. Mirror? What you would expect, 55MB/s sustained writes, only 57MB/s sustained reads.

B&W G3 (G4/500). Adaptec 39160. SoftRAID 2.2.2 (2) 2G X15 Cheetah
(all in MB/s with ATTO ExpressTools 2.7, 8MB Max Transfer, Sampe: 2)
PR: 107 SR: 105
PW: 129 SW: 115

What is also interesting is that after filling the volume(s) to 40-50% there was a 20% hit in performance.
SR: 93 SW: 98

Two 10k drives on the other channel, and they get 80MB/s sustained read, 100MB/s sustrained writes. Interesting.

I need to clean it up. Would like to know how you post a pict into a message.

I'm happy with the investment in the Adapatec 39160 card, even if OS X support is missing (so is SoftRAID for X, along with Disk Warrior - stuff we may have to wait until after the Jaguar version has shipped).

Gregory

[This message has been edited by Gregory (edited 11 May 2002).]

ricks
05-10-2002, 08:08 PM
Gregory,

I'll give that a shot with the 4 AtlasIIIs and the UL3D. Be fun to see what the limit is.

Do you have any idea what we give up with the Radeon card in a 33 mhz slot? I'd not be satisfied with the graphics being way subpar.

Posting a jpeg is done by dropping the photo on my homepage at mac.com and then linking it via UBB code. The code is in the faq.

The homepage is free at mac.com. The address you use if you put the picture in your Public folder is: http://homepage.mac.com/your user name/.Public/picture name

Hope that's what you wanted.

Rick

TZ
05-11-2002, 09:18 AM
Whatever is lost to the Radeon can't be much for non-games but you run PS. then again, getting numbers like 115MB/s really help. I don't think even heavy web browsing with 50 windows open puts a load on Radeon (millions, 1280x1024).

With both drives on same channel, mirror SW holds at 54MB/s - next is to see what happens on separate channels for the RAID and mirror.

Gregory

Doc
05-13-2002, 12:36 PM
I'm just about to hook up my hardware RAID to a UL3D in my 533DA Mac OS X server .. we'll see how well that works out - and compare transfer rates against the Miles2 in my G3 upgraded PTP system..


Chris

TZ
05-14-2002, 11:49 AM
I know there is overhead using both channels to create raid volumes (both mirror and stripe) but was horrified to find that a 2-drive X15 raid on 39160 shows huge punishment over putting the raid on one channel. maybe with larger number of drives it isn't so substantial, but drop off from 115MB/s SW down to 75MB/s for stripe, and from 54MB/s down to 39MB/s for mirror - just by using two channel rather than one channel.

Don't know that anyone would want to or try to, but I wasn't expecting such a huge performance hit.

One thing I like with Power Domain Control is that while ExpressTools from ATTO will show what mode and whether a drive is running at 80 MHz (double click on the drive in the list) or what the controller's settings on), it is much easier to see and monitor. Found that one terminator may not have been seated properly, or there was dust hair, something causing it to fall to narrow 40 MHz mode rather than wide AND operate at 80 MHz which is needed for Ultra160.

if the 29160 works properly in 66 MHz slots, might have saved a little using that instead of 39160, but it definitely outperforms UL3S and probably UL3D - unless those are 66 MHz version of the card (which ATTO says should NOT be used in 33 MHz slots, meaning that they are not true dual function cards).

Can't use this RAID under OS X, SoftRAID will have different drivers for each OS, and Adaptec says that their cards do not currently support Apple's OS X RAID (meaning that ATTO's ExpressStripe is the only one that may work AND work in both OS 9 and OS X) which is an interesting turn of events.

Gregory

kaye
05-14-2002, 12:43 PM
I had the opposite experience (see the RAID Data Base).

Even with the additional overhead of using both channels to create stripe-only raid volumes, my 2-drive G2 X15 raid on a UL3D shows a slight improvement in sustained writes over putting the raid on one channel.

2-drive, 2-channel SW 118MB/s
2-drive, 1-channel SW 116MB/s

The dual channel also yielded much higher Peak Read/Write.

In your case, could it be due to the additional overhead of also mirroring? I don't get it. k

TZ
05-14-2002, 02:28 PM
kaye,

Maybe it is the external cables (one granite cable failed and they do tend to separate, having had it happen twice) or something else. Or a flaw in 39160. Maybe ATTO stuffs more SDRAM on their board? Are you using G4? B&W? 2MB/s is ~1.5% difference.

Remember I was stuck, hitting a 53-59MB/s solid WALL on UL3S and B&W - using any slot didn't make a bit of difference or help. Or using two Ultra2 cards. Even Rick with UL3D and B&W saw the same effect.

So for me, this 'works' and is a huge 100+% improvement. Maybe the fact that I have one internal RAID using 10k Cheetahs and this is external RAID using X15's - even though only testing - shouldn't matter or cause "noise" or degredation.

I'll have to retest again after I make some changes.

Gregory

kaye
05-14-2002, 04:19 PM
Gregory,

Using a G4-800DP. Yes, just a slight improvement but no drop like you experienced. k