RE: performance limitations of linux raid
(I really hate how Outlook makes you answer in FRONT of the message, what a dumb design...) Well, without spending the time I should thinking about my answer, I'll say there are many things which impact performance, most of which we've seen talked about here: 1 - how fast can you get data off the media? 2 - related - does the data just happen to be in drive cache? 3 - how fast can you get data from the drive to the controller? 4 - how fast can you get data from the controller into system RAM? 5 - how fast can you get that data to the user? (assuming reads - writes are similar, but reversed - for the most part) Number 1 relates to rotational delay, seek times, and other things I'm probably forgetting. (Like sector skewing (is that the right term? I forget!) - where you try to put the 'next' sector where its ready to be read by the head (on a sequential read) just after the system has gotten around to asking for that sector (boy, I can see the flames coming already! ;-) / 2 Number 2 relates to how smart your drive is, and too smart a drive can actually slow you down by being in the wrong place reading data you don't want when you go ask it for data somewhere else. Number 3 relates to not only the obvious issue of how fast the scsi bus is, but how congested it is. If you have 15 devices which can sustain a data rate (including rotational delays and cache hits) of 10 megabytes/sec, and your scsi bus can only pass 20 MB/sec, then you should not put more than 2 of those devices on that bus - thus requiring more and more controllers... (and I'm ignoring any issues of contention, as I'm not familiar with the low level of scsi enough to know about it) Number 4 relates to your system bus bandwidth, DMA speed, system bus loading, etc. Number 5 relates to how fast your cpu is, how well written the driver is, and other things I'm probably forgetting. (Like , can the OS actually HANDLE 2 things going on at once - and do floppy accesses take priority over later requests for hard disk accesses?) So maximizing performance is not a 1-variable exercise. And you don't always have the control you'd like over all the variables. And paying too much attention to only one while ignoring others can easily cause you to make really silly statements like: "Wow, I've really got a fast system here - I have an ULTRA DMA 66 drive on my P133 - really screams. And with that nice new 199x cdrom drive with it as secondary - wow, I really SCREAM through those CD's!" Um, well, sure, uh-huh. Most of you on this list see the obvious errors there - I've seen some pretty smart people do similar things (but not so obvious to most folks) by missing some of the above issues. Well, this is way longer than I expected, so I'll quit before I get into any MORE trouble than I probably am already! rusty -Original Message- From: Bob Gustafson [mailto:[EMAIL PROTECTED]] Sent: Thursday, May 04, 2000 5:18 PM To: [EMAIL PROTECTED]; [EMAIL PROTECTED] Subject: Re: performance limitations of linux raid I think the original answer was more to the point of Performance Limitation. The mechanical delays inherent in the disk rotation are much slower than the electronic or optical speeds in the connection between disk and computer. If you had a huge bank of semiconductor memory, or a huge cache or buffer which was really helping (i.e., you wanted the information that was already in the cache or buffer), then things get more complicated. BobG
RE: performance limitations of linux raid
From: Gregory Leblanc [mailto:[EMAIL PROTECTED]] ..., that would suck up a lot more host CPU processing power than the 3 SCSI channels that you'd need to get 12 drives and avoid bus saturation. not to mention the obvious bus slot loading problem ;-) rc
Re: performance limitations of linux raid
Not entirely, there is a fair bit more CPU overhead running an IDE bus than a proper SCSI one. A "fair" bit on a 500mhz+ processor is really negligible. Ehem, a fair bit on a 500Mhz CPU is ~ 30%. I have watched a *single* UDMA66 drive (with read ahead, multiblock io, 32bit mode, and dma transfers enabled) on a 2.2.14 + IDE + RAID patched take over 30% of the CPU during disk activity. The same system with a 4 x 28G RAID0 set running would be .1% idle during large copies. An exactly configured system with UltraWide SCSI instead of IDE sits ~ 95% idle during the same ops. Try turning on DMA
Re: performance limitations of linux raid
On Fri, 5 May 2000, Michael Robinton wrote: Not entirely, there is a fair bit more CPU overhead running an IDE bus than a proper SCSI one. A "fair" bit on a 500mhz+ processor is really negligible. Ehem, a fair bit on a 500Mhz CPU is ~ 30%. I have watched a *single* UDMA66 drive (with read ahead, multiblock io, 32bit mode, and dma transfers enabled) on a 2.2.14 + IDE + RAID patched take over 30% of the CPU during disk activity. The same system with a 4 x 28G RAID0 set running would be .1% idle during large copies. An exactly configured system with UltraWide SCSI instead of IDE sits ~ 95% idle during the same ops. Try turning on DMA Ahem, try re-reading the above, line 3 first word! --- As folks might have suspected, not much survives except roaches, and they don't carry large enough packets fast enough... --About the Internet and nuclear war.
RE: performance limitations of linux raid
The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, ... Well, yeah, and so whatever happened to optical scsi? I heard that you could get 1 gbit/sec (or maybe gByte?) xfer, and you could go 1000 meters - or is this not coming down the pike? (optical scsi - meaning using fiber instead of ribbon cable to interconnect controller to drive) rusty
Re: performance limitations of linux raid
On Thu, May 04, 2000 at 08:35:52AM -0700, Carruth, Rusty wrote: The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, ... Well, yeah, and so whatever happened to optical scsi? I heard that you could get 1 gbit/sec (or maybe gByte?) xfer, and you could go 1000 meters - or is this not coming down the pike? (optical scsi - meaning using fiber instead of ribbon cable to interconnect controller to drive) Optical is an interesting idea. Fiber has a better throughput (I think the limiting factor is the LED/laser modulation), but fiber has a longer latency. I.e., light travels slower in fiber than electricity through wire. (No, light doesn't travel at 'light speed' through a dense medium like plastic or glass.) But you can pack more data in a fiber. (I.e., more on-off transitions per second.) I think fiber is mostly used for running long distances, or for isolating (electricly) between nodes/networks. Some NSP's require that your link be fiber to prevent the possibility of you frying their equipment if you, say, wire your DS1 (T1) pairs to an electrical outlet.. ;') For shorter distances, I think differential pairs of wires is faster/cheaper/easier for most things. Phil -- Philip Edelbrock -- IS Manager -- Edge Design, Corvallis, OR [EMAIL PROTECTED] -- http://www.netroedge.com/~phil PGP F16: 01 D2 FD 01 B5 46 F4 F0 3A 8B 9D 7E 14 7F FB 7A
RE: performance limitations of linux raid
-Original Message- From: Carruth, Rusty [mailto:[EMAIL PROTECTED]] Sent: Thursday, May 04, 2000 8:36 AM To: [EMAIL PROTECTED] Subject: RE: performance limitations of linux raid The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, ... Well, yeah, and so whatever happened to optical scsi? I heard that you could get 1 gbit/sec (or maybe gByte?) xfer, and you could go 1000 meters - or is this not coming down the pike? 1 Gbit/sec is approximately equal to 120Mbytes/sec. Ultra160 SCSI is faster than that. (optical scsi - meaning using fiber instead of ribbon cable to interconnect controller to drive) Fiber Channel, or SCSI over Fiber Channel? Sun has an FC array that allows you to have 500metres between the server and the array, at 100MegaBytes/sec data transfer. I don't know what they're going to do with FC, but it's been outpaced for speed by SCSI. Greg
Re: performance limitations of linux raid
On Wed, 3 May 2000, Michael Robinton wrote: The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, the big IBM drives (20 - 40 gigs) have a limitation of about 27mbs for both the 7200 and 10k rpm models. The Drives to come will have to make trade-offs between density and speed, as the technology's in the works have upper constraints on one or the other. So... given enough controllers (either scsii on disk or ide individual) the limit will be related to the bandwidth of the disk interface rather than the speed of the processor it's talking too. Not entirely, there is a fair bit more CPU overhead running an IDE bus than a proper SCSI one. --- As folks might have suspected, not much survives except roaches, and they don't carry large enough packets fast enough... --About the Internet and nuclear war.
Re: performance limitations of linux raid
On Thu, 4 May 2000, Christopher E. Brown wrote: On Wed, 3 May 2000, Michael Robinton wrote: The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, the big IBM drives (20 - 40 gigs) have a limitation of about 27mbs for both the 7200 and 10k rpm models. The Drives to come will have to make trade-offs between density and speed, as the technology's in the works have upper constraints on one or the other. So... given enough controllers (either scsii on disk or ide individual) the limit will be related to the bandwidth of the disk interface rather than the speed of the processor it's talking too. Not entirely, there is a fair bit more CPU overhead running an IDE bus than a proper SCSI one. A "fair" bit on a 500mhz+ processor is really negligible.
Re: performance limitations of linux raid
On Sun, 23 Apr 2000, Chris Mauritz wrote: I wonder what the fastest speed any linux software raid has gotten, it would be great if the limitation was a hardware limitation i.e. cpu, (scsi/ide) interface speed, number of (scsi/ide) interfaces, drive speed. It would be interesting to see how close software raid could get to its hardware limitations. RAID0 seemed to scale rather linearly. I don't think there would be much of a problem getting over 100mbits/sec on an array of 8-10 ultra2 wide drives. I ultimately stopped fiddling with software RAID on my production boxes as I needed something that would reliably do hot swapping of dead drives. So I've switched to using Mylex ExtremeRAID 1100 cards instead (certainly not the card you want to use for low budget applications...heh). Umm, I can get 13,000K/sec to/from ext2 from a *single* UltraWide Cheeta (best case, *long* reads, no seeks). 100Mbit is only 12,500K/sec. A 4 drive UltraWide Cheeta array will top out an UltraWide bus at 40MByte/sec, over 3 times the max rate of a 100Mbit ethernet. --- As folks might have suspected, not much survives except roaches, and they don't carry large enough packets fast enough... --About the Internet and nuclear war.
Re: performance limitations of linux raid
From [EMAIL PROTECTED] Wed May 3 20:38:05 2000 Umm, I can get 13,000K/sec to/from ext2 from a *single* UltraWide Cheeta (best case, *long* reads, no seeks). 100Mbit is only 12,500K/sec. A 4 drive UltraWide Cheeta array will top out an UltraWide bus at 40MByte/sec, over 3 times the max rate of a 100Mbit ethernet. Yup, that's why all my RAID boxen use gigabit ether cards. 8-) The Netgear cards are only about $300. The switches are a bit expensive though. 8-) Cheers, Chris -- Christopher Mauritz [EMAIL PROTECTED]
Re: performance limitations of linux raid
The primary limitation is probably the rotational speed of the disks and how fast you can rip data off the drives. For instance, the big IBM drives (20 - 40 gigs) have a limitation of about 27mbs for both the 7200 and 10k rpm models. The Drives to come will have to make trade-offs between density and speed, as the technologys in the works have upper constraints on one or the other. So... given enough controllers (either scsii on disk or ide individual) the limit will be related to the bandwidth of the disk interface rather than the speed of the processor it's talking too. On Wed, 3 May 2000, Christopher E. Brown wrote: On Sun, 23 Apr 2000, Chris Mauritz wrote: I wonder what the fastest speed any linux software raid has gotten, it would be great if the limitation was a hardware limitation i.e. cpu, (scsi/ide) interface speed, number of (scsi/ide) interfaces, drive speed. It would be interesting to see how close software raid could get to its hardware limitations. RAID0 seemed to scale rather linearly. I don't think there would be much of a problem getting over 100mbits/sec on an array of 8-10 ultra2 wide drives. I ultimately stopped fiddling with software RAID on my production boxes as I needed something that would reliably do hot swapping of dead drives. So I've switched to using Mylex ExtremeRAID 1100 cards instead (certainly not the card you want to use for low budget applications...heh). Umm, I can get 13,000K/sec to/from ext2 from a *single* UltraWide Cheeta (best case, *long* reads, no seeks). 100Mbit is only 12,500K/sec. A 4 drive UltraWide Cheeta array will top out an UltraWide bus at 40MByte/sec, over 3 times the max rate of a 100Mbit ethernet. --- As folks might have suspected, not much survives except roaches, and they don't carry large enough packets fast enough... --About the Internet and nuclear war.
: Re: performance limitations of linux raid
If you are looking for one of the higest performance systems we have ever seen, visit www.raidzone.com. These are real systems. John ___ Are you a Techie? Get Your Free Tech Email Address Now! Many to choose from! Visit http://www.TechEmail.com
Re: performance limitations of linux raid
More notes on the 8 IDE drive raid5 system I built, and the 3ware controller. Edwin Hakkennes wrote: May I ask how these ide-ports and the attatched disks show up under Redhat 6.2? Are they just standard ATA33 or ATA66 controllers which can be used in software raid? Or is only the sum of the attatched disks usable as one scsi disk? The 3-ware site doesn't mention anything like this. Did you need any quirks to get this running? The 3ware controller is supported in the latest 2.2.X kernels. I used 2.2.15pre17. It is listed with the scsi modules and indeed, that is how the drives show up. In my case, I had an 9.1GB LVD drive on a tekram dc390u2 controller as my system drive (sda) so the four Maxtor DMA66 drives became sdb - sde. I didn't use 3ware bios or setup at all. 3ware expects you to use the controller as advertised - with their setup software to create a raid0, raid1 or raid1-0 array. I haven't tried that yet. I just wanted a lot of fast dma66 controllers in a single slot. I figured out how to address the individual drives, which as you can see, was pretty straight forward, and let the linux software handle the raid. 3ware doesn't mention anything like this, because it is news to them. They have no experience using their boards with software raid or raid5. Unfortunatedly, their boards are somewhat pricey, reflecting I suppose, there status as raid controllers, as opposed to UDMA66 controllers. At Linux Beach we have just become dealers for 3ware and can offer these boards at slightly less than the 3ware prices: QTY ITEM NO. DESCRIPTION PRICE -- RAID-3W5200L 3WARE ESCALADE 2PORT IDE RAID CONTROLLER $119.00 RAID-3W5400L 3WARE ESCALADE 4PORT IDE RAID CONTROLLER $219.00 RAID-3W5800L 3WARE ESCALADE 8PORT IDE RAID CONTROLLER $419.00 So my question basically is: Could you send me the output of cat /proc/pci cat /proc/partitions cat /proc/mdstat ls -laR /proc/ide This system currently a production server at my client's site so I don't have easy access to it. So I can't send you the proc outputs you requested. However they have promised to let me have it for awhile at sometime in the future for futher benchmarking and tuning. Hopefully I can get it then. Here is the raidtab: /etc/raidtab raiddev /dev/md0 raid-level 5 nr-raid-disks 8 nr-spare-disks 0 persistent-superblock 1 chunk-size 128 parity-algorithm left-symmetric device /dev/hda1 raid-disk 0 device /dev/hde1 raid-disk 1 device /dev/hdc1 raid-disk 2 device /dev/hdg1 raid-disk 3 device /dev/sdb1 raid-disk 4 device /dev/sdc1 raid-disk 5 device /dev/sdd1 raid-disk 6 device /dev/sde1 raid-disk 7 Benno Senoner [EMAIL PROTECTED]> wrote: I went to the 3WARE site. That is http://www.3ware.com really nice the 8 IDE channel version and cheap too. :-) I noticed that they do not support UDMA/66. (at least the PDF says so). On second look, this is correct. I know they used UDMA66 drives in their benchmarks: http://3ware.com/products/3wareEscaladePerformanceAnalysis_kc.pdf and recommend UDMA66 drives for best preformance. Do you think that the impact is negligible when there is only one drive per chanel ?(At least I think that there are not that many EIDE disks which can sustain 33MB/sec all the time). The load (in terms of bandwidth) generated on the bus/CPU by multiple (8) disks is quite high IMHO, therefore 66MB/sec * 8 (even if every drive would be able to deliver that kind of bandwidth) , would likely saturate your mobo/CPU, shifting the bottleneck from the disks to the memory/CPU subsystem. The controller ships with single drive 80 wire cables, and the numbers I got were equivalent to those I got from the other DMA66 controllers running master only. BTW, do you need a 2.3.x to work with these 3WARE monsters ? Or are there patches backported to 2.2.x floating around ? I know it's in 2.2.15pre14 and above and modules are available at the 3ware site for RH6.1, RH6.2 and Suse 6.3. TTFN Clay J. Claiborne, Jr., President Cosmos Engineering Company 1550 South Dunsmuir Ave. Los Angeles, CA 90019 (323) 930-2540 (323) 930-1393 Fax http:www.CosmosEng.com Email: [EMAIL PROTECTED]
Re: performance limitations of linux raid
In the last 24 hours ive been getting them when e2fsck runs after rebooting. Usual cause of rebooting is irq causeing lockup, or endlessly trying looping trying to get an irq. Im convinced its my hpt366 controller, ive mentioned my problem in a few channels, no luck yet. I used to think it was the raid code, but i get it with lvm as well, it happens more often from reading than writting via the HPT366, the more load placed on the controller the more likely it is to lockup, one drive by itself it just losses interrupts (sometimes it can recover), if use three or four channels, using both my onboard hpt366 and my pci card it locks up hard in a fraction of a second I want to try and work this problem out, im not a kernel hacker though. me too ... have the same problem over here with hpt366 (abit hotrod) and 4 ibm 25gb disks. i thought this happens when array i/o is high, but recently seen some (with successfull recovery) when disks were quiet. i guess we really need some expert kernel hacker to sort this out... btw, i'm new to this list and have some (already answered i guess) questions. what's the current status of linux raid? 2.2 patch seems to be stuck at .11. is there any general recipie to make it work in recen kernels? i need to have it working in 2.2.15pre19 with ide patch also. please enlighten me :) |Pegasus|
Re: performance limitations of linux raid
The coolest guy you know wrote: Clay Claiborne wrote: For what its worth, we recently built an 8 ide drive 280GB raid5 system. Benchmarking with HDBENCH we got 35.7MB/sec read and 29.87MB/sec write. With DBENCH and 1 client we got 44.5 MB/sec with 3 clients it dropped down to about 43MB/sec. The system is a 600Mhz P-3 on a ASUS P3C2000 with 256MB of ram, the raid drives are 40GB Maxtor DMA66, 7200 RPM, and each is run as master on its own channel. Turning on DMA seems to be the key. Benchmarking the individual drives with HDBENCH we got numbers like 2.57MB/sec read and 3.27MB/sec write with DMA off and it jumped up to 24.7MB/sec read and 24.2MB/sec write with it on. That, and enough processing power to see that paritity calc is not a bottleneck. -- Clay J. Claiborne, Jr., President Cosmos Engineering Company 1550 South Dunsmuir Ave. Los Angeles, CA 90019 (323) 930-2540 (323) 930-1393 Fax http:www.CosmosEng.com Email: [EMAIL PROTECTED] You have some pretty impressive numbers there. What controllers did you use? Chris -- 2 Masters from the ASUS P3C2000 CDU Board, 2 Masters on a CMD648 based PCI controller and 4 Masters on a 3Wave 4 port RAID Controller. I don't use the 3Wave as a RAID controller though, just as a very good 4 channel Ultra66 board. They also make an 8 channel board, and that's what I'm going to use next time. I can put two in a system an mount 16 IDE drives for raid while saving the on board IDE's for the usual stuff. Clay. Cosmos Engineering Company 1550 South Dunsmuir Ave. Los Angeles, CA 90019 (323) 930-2540 (323) 930-1393 Fax http:www.CosmosEng.com Email: [EMAIL PROTECTED]
Re: performance limitations of linux raid
2 Masters from the ASUS P3C2000 CDU Board, 2 Masters on a CMD648 based PCI controller and 4 Masters on a 3Wave 4 port RAID Controller. I don't use the 3Wave as a RAID controller though, just as a very good 4 channel Ultra66 board. They also make an 8 channel board, and that's what I'm going to use next time. I can put two in a system an mount 16 IDE drives for raid while saving the on board IDE's for the usual stuff. So do the standard IDE drivers in the kernel detect and handle all these master controllers?? or is a special driver required? Very Interested!! Michael [EMAIL PROTECTED]
Re: performance limitations of linux raid
bug1 wrote: Clay Claiborne wrote: For what its worth, we recently built an 8 ide drive 280GB raid5 system. Benchmarking with HDBENCH we got 35.7MB/sec read and 29.87MB/sec write. With DBENCH and 1 client we got 44.5 MB/sec with 3 clients it dropped down to about 43MB/sec. The system is a 600Mhz P-3 on a ASUS P3C2000 with 256MB of ram, the raid drives are 40GB Maxtor DMA66, 7200 RPM, and each is run as master on its own channel. Turning on DMA seems to be the key. Benchmarking the individual drives with HDBENCH we got numbers like 2.57MB/sec read and 3.27MB/sec write with DMA off and it jumped up to 24.7MB/sec read and 24.2MB/sec write with it on. That, and enough processing power to see that paritity calc is not a bottleneck. Can you use your raid system with DMA turned on or do you get irq timouts like me ? gonna jump on the bandwagon here...:-) How often do you get them?? On all the disks, or just some of them. In all the modes or just with ultra66??? i get some every couple of days, normally resulting in the dma getting disabled for the specific drive...:-( but the filesystems seem to be ok... remo
Re: performance limitations of linux raid
On Mon, 24 Apr 2000, Frank Joerdens wrote: I've been toying with the idea of getting one of those for a while, but there doesn't seem to be a linux driver for the FastTrack66 (the RAID card), only for the Ultra66 (the not-hacked IDE controller), and that driver has only 'Experimental' status with current production kernels: Clue: the Promise IDE RAID controller is NOT a hardware RAID controller. Promise IDE RAID == Software RAID where the software is written by Promise and sitting on the ROM on the Promise card getting called by the BIOS. I even wrote to Promise to ask when or if a linux driver might become available, but didn't get much of an answer (they replied that there was a driver available for the Ultra, although I had specifically asked for the RAID card). Dont bother... A driver for Promise RAID == Software RAID. You already ahve software RAID in linux If anyone hears about a Linux driver for this card, I'd like to know. Cheers, Frank -- Paul Jakma [EMAIL PROTECTED] PGP5 key: http://www.clubi.ie/jakma/publickey.txt --- Fortune: I use not only all the brains I have, but all those I can borrow as well. -- Woodrow Wilson
Re: performance limitations of linux raid
On Tue, Apr 25, 2000 at 10:28:46PM +0100, Paul Jakma wrote: Clue: the Promise IDE RAID controller is NOT a hardware RAID controller. Promise IDE RAID == Software RAID where the software is written by Promise and sitting on the ROM on the Promise card getting called by the BIOS. Clue: this is the way every RAID controller I know of works these days. PS: Linux doesn't use BIOS to access devices.
RE: performance limitations of linux raid
-Original Message- From: Daniel Roesen [mailto:[EMAIL PROTECTED]] Sent: Tuesday, April 25, 2000 3:07 PM To: [EMAIL PROTECTED] Subject: Re: performance limitations of linux raid On Tue, Apr 25, 2000 at 10:28:46PM +0100, Paul Jakma wrote: Clue: the Promise IDE RAID controller is NOT a hardware RAID controller. Promise IDE RAID == Software RAID where the software is written by Promise and sitting on the ROM on the Promise card getting called by the BIOS. Clue: this is the way every RAID controller I know of works these days. Then you've never used a RAID card. I've got a number of RAID cards here, 2 from compaq, 1 from DPT, and another from HP (really AMI), and all of them implement RAID functions like striping, double writes (mirroring), and parity calculations for RAID4/5 in firmware, using an onboard CPU. All the controllers here are i960 based, but I've heard that the StrongARM procs are much faster at parity caclculations. The controllers that I've used that are software are the Adaptec AAA series boards. The other one that I know of is this Promise thing. Greg
Re: performance limitations of linux raid
On Mon, Apr 24, 2000 at 09:13:20PM -0400, Scott M. Ransom wrote: Then I moved back to kernel 2.2.15-pre18 with the RAID and IDE patches and here are my results: RAID0 on Promise Card 2.2.15-pre18 (1200MB test) -- ---Sequential Output ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 6833 99.2 42532 44.4 18397 42.2 7227 98.3 47754 33.0 182.8 1.5 ** When doing _actual_ work (I/O bound reads on huge data sets), I often see sustained read performance as high as 50MB/s. Tests on the individual drives show 28+ MB/s. What stripe size, CPU and memory is used here? I have a similar setup (2.2.15pre19, IDE+RAID patches), 4 master IDE Deskstars, on board VIA and offboard Promise controllers, and K6-2/500 256M ram and see 22MB/s native but a max of 28 MB/s for all stripes. dd/hdparm -t on multiple drives simultaneously appears to show complete contention between the separate chains. hdparm -t /dev/hde 19.16 MB/sec ( hdparm -t /dev/hde ) ; (hdparm -t /dev/hdg ) 11.43 MB/sec 10.47 MB/sec With all four drives throughput per drive is less than 7MB/sec. With RAID0 across all four drives I get 28 MB/sec according to bonnie, vs 22 MB/sec on single drives. I had been attributing that to a singly-entrant IDE driver in 2.2, but your results make me think there's some other reason I don't see linear speedups. Is this a dual CPU system perhaps? Something unusual about the interrupt handling? UDMA33 vs. UDMA 66 (I'm using 40 conductor cables, perhaps I need the 80s)?
Re: performance limitations of linux raid
remo strotkamp wrote: bug1 wrote: Clay Claiborne wrote: For what its worth, we recently built an 8 ide drive 280GB raid5 system. Benchmarking with HDBENCH we got 35.7MB/sec read and 29.87MB/sec write. With DBENCH and 1 client we got 44.5 MB/sec with 3 clients it dropped down to about 43MB/sec. The system is a 600Mhz P-3 on a ASUS P3C2000 with 256MB of ram, the raid drives are 40GB Maxtor DMA66, 7200 RPM, and each is run as master on its own channel. Turning on DMA seems to be the key. Benchmarking the individual drives with HDBENCH we got numbers like 2.57MB/sec read and 3.27MB/sec write with DMA off and it jumped up to 24.7MB/sec read and 24.2MB/sec write with it on. That, and enough processing power to see that paritity calc is not a bottleneck. Can you use your raid system with DMA turned on or do you get irq timouts like me ? gonna jump on the bandwagon here...:-) How often do you get them?? On all the disks, or just some of them. In all the modes or just with ultra66??? i get some every couple of days, normally resulting in the dma getting disabled for the specific drive...:-( but the filesystems seem to be ok... remo Yea, i get them on my disks (Quantum XA and KX, both my IBM DPTA 372050), they usually start at the last ide channel and work backwards. In the last 24 hours ive been getting them when e2fsck runs after rebooting. Usual cause of rebooting is irq causeing lockup, or endlessly trying looping trying to get an irq. Im convinced its my hpt366 controller, ive mentioned my problem in a few channels, no luck yet. I used to think it was the raid code, but i get it with lvm as well, it happens more often from reading than writting via the HPT366, the more load placed on the controller the more likely it is to lockup, one drive by itself it just losses interrupts (sometimes it can recover), if use three or four channels, using both my onboard hpt366 and my pci card it locks up hard in a fraction of a second I want to try and work this problem out, im not a kernel hacker though. Anyone have any advice on how to get into kernel debugging, i know c, i dont know the kernel though. I know how to use ksymoops thats about it. Glenn
Re: performance limitations of linux raid
Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit What stripe size, CPU and memory is used here? System is a dual-cpu PII 450Mhz with 256MB RAM. Disks are configured with chunk-size of 32kb (ext2 block-size is 4kb). Is this a dual CPU system perhaps? Something Yes. See above. unusual about the interrupt handling? UDMA33 vs. UDMA 66 (I'm using 40 conductor cables, perhaps I need the 80s)? I am using UDMA66. When testing against UDMA33, I found a 10-15% speed difference. To get that speed increase you must have the UDMA66 cables... Scott -- Scott M. Ransom Phone: (781) 320-9867 Address: 75 Sanderson Ave. email: [EMAIL PROTECTED] Dedham, MA 02026 PGP Fingerprint: D2 0E D0 10 CD 95 06 DA EF 78 FE 2B CB 3A D3 53
Re: performance limitations of linux raid
I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Check http://www.tomshardware.com , they review the Promise IDE RAID card (the hacked one for $30). They get some pretty insane throughputs on some ATA66 drives. Gee, they look pretty awesome, excellent performance. The only drawback i can think of is that it wouldnt be as flexible as software raid. I've been toying with the idea of getting one of those for a while, but there doesn't seem to be a linux driver for the FastTrack66 (the RAID card), only for the Ultra66 (the not-hacked IDE controller), and that driver has only 'Experimental' status with current production kernels: - snip --- CONFIG_BLK_DEV_PDC4030: This driver provides support for the secondary IDE interface and cache of Promise IDE chipsets, e.g. DC4030 and DC5030. This driver is known to incur timeouts/retries during heavy I/O to drives attached to the secondary interface. CDROM and TAPE devices are not supported yet. This driver is enabled at runtime using the "ide0=dc4030" kernel boot parameter. See the Documentation/ide.txt and drivers/block/pdc4030.c files for more info. - snap --- I even wrote to Promise to ask when or if a linux driver might become available, but didn't get much of an answer (they replied that there was a driver available for the Ultra, although I had specifically asked for the RAID card). If anyone hears about a Linux driver for this card, I'd like to know. Cheers, Frank -- frank joerdens joerdens new media heinrich-roller str. 16/17 10405 berlin germany e: [EMAIL PROTECTED] t: +49 (0)30 44055471 f: +49 (0)30 44055475 h: http://www.joerdens.de pgp public key: http://www.joerdens.de/pgp/frank_joerdens.asc
Re: performance limitations of linux raid
I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Sure it is. go to the ibm site and look at the specs on all the new high capacity drives. Without regard to the RPM, they are all spec'd to rip data off the drive at around 27mb/sec continuous. [EMAIL PROTECTED]
Re: performance limitations of linux raid
There's "specs" and then there's real life. I have never seen a hard drive that could do this. I've got brand new IBM 7200rpm ATA66 drives and I can't seem to get them to do much better than 6-7mb/sec with either Win98, Win2000, or Linux. That's with Abit BH6, an Asus P3C2000, and Supermicro PIIIDME boards. And yes, I'm using an 80 conductor cable. I'm using Wintune on the windows platforms and bonnie on Linux to do benchmarks. Cheers, Chris - Original Message - From: "Michael" [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Monday, April 24, 2000 5:10 PM Subject: Re: performance limitations of linux raid I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Sure it is. go to the ibm site and look at the specs on all the new high capacity drives. Without regard to the RPM, they are all spec'd to rip data off the drive at around 27mb/sec continuous. [EMAIL PROTECTED]
Re: performance limitations of linux raid
There's "specs" and then there's real life. I have never seen a hard drive that could do this. I've got brand new IBM 7200rpm ATA66 drives and I can't seem to get them to do much better than 6-7mb/sec with either Win98, Win2000, or Linux. That's with Abit BH6, an Asus P3C2000, and Supermicro PIIIDME boards. And yes, I'm using an 80 conductor cable. I'm using Wintune on the windows platforms and bonnie on Linux to do benchmarks. turn udma modes on in the bios and run hdparm -d 1 /dev/hda (where hda == drive device) the re-run your specs I think you'll find the speed is stepped up dramatically. i'm getting 16MB/s write and 22MB/s read on the same drive. I got for crap w/o the dma turned on via hdparm -sv
RE: performance limitations of linux raid
-Original Message- From: Chris Mauritz [mailto:[EMAIL PROTECTED]] Sent: Monday, April 24, 2000 2:30 PM To: [EMAIL PROTECTED]; [EMAIL PROTECTED] Subject: Re: performance limitations of linux raid There's "specs" and then there's real life. I have never seen a hard drive that could do this. I've got brand new IBM 7200rpm ATA66 drives and I can't seem to get them to do much better than 6-7mb/sec with either Win98, Win2000, or Linux. That's with Abit BH6, an Asus P3C2000, and Supermicro PIIIDME boards. And yes, I'm using an 80 conductor cable. I'm using Wintune on the windows platforms and bonnie on Linux to do benchmarks. I don't believe the specs either, because they are for the "ideal" case. However, I think that either your benchmark is flawed, or you've got a crappy controller. I have a (I think) 5400 RPM 4.5GB IBM SCA SCSI drive in a machine at home, and I can easily read at 7MB/sec from it under Solaris. Linux is slower, but that's because of the drivers for the SCSI controller. I haven't done any benchmarks on my IDE drives because I already know that they're SLOW. Greg Cheers, Chris - Original Message - From: "Michael" [EMAIL PROTECTED] To: [EMAIL PROTECTED] Sent: Monday, April 24, 2000 5:10 PM Subject: Re: performance limitations of linux raid I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Sure it is. go to the ibm site and look at the specs on all the new high capacity drives. Without regard to the RPM, they are all spec'd to rip data off the drive at around 27mb/sec continuous. [EMAIL PROTECTED]
Re: performance limitations of linux raid
There's "specs" and then there's real life. I have never seen a hard drive that could do this. I've got brand new IBM 7200rpm ATA66 drives and I can't seem to get them to do much better than 6-7mb/sec with either Win98, Win2000, or Linux. That's with Abit BH6, an Asus P3C2000, and Supermicro PIIIDME boards. And yes, I'm using an 80 conductor cable. I'm using Wintune on the windows platforms and bonnie on Linux to do benchmarks. turn udma modes on in the bios and run hdparm -d 1 /dev/hda (where hda == drive device) just FYI, 3 disk raid 5 on udma33 asus mother board + 1 udma33 promise controller. disks are maxtor 87000D8's speced at a maximum of media to interface transfer rate of 14.7 mbs -- doesn't include any seek time. my box has 128megs so ran a 500 meg bonnie. results block write = 7.9mbs block read = 10.5 mps pretty good for an old 166 K6 Michael [EMAIL PROTECTED]
Re: performance limitations of linux raid
For what its worth, we recently built an 8 ide drive 280GB raid5 system. Benchmarking with HDBENCH we got 35.7MB/sec read and 29.87MB/sec write. With DBENCH and 1 client we got 44.5 MB/sec with 3 clients it dropped down to about 43MB/sec. The system is a 600Mhz P-3 on a ASUS P3C2000 with 256MB of ram, the raid drives are 40GB Maxtor DMA66, 7200 RPM, and each is run as master on its own channel. Turning on DMA seems to be the key. Benchmarking the individual drives with HDBENCH we got numbers like 2.57MB/sec read and 3.27MB/sec write with DMA off and it jumped up to 24.7MB/sec read and 24.2MB/sec write with it on. That, and enough processing power to see that paritity calc is not a bottleneck. -- Clay J. Claiborne, Jr., President Cosmos Engineering Company 1550 South Dunsmuir Ave. Los Angeles, CA 90019 (323) 930-2540 (323) 930-1393 Fax http:www.CosmosEng.com Email: [EMAIL PROTECTED]
RE: performance limitations of linux raid
-Original Message- From: Scott M. Ransom [mailto:[EMAIL PROTECTED]] Sent: Monday, April 24, 2000 6:13 PM To: [EMAIL PROTECTED] Cc: [EMAIL PROTECTED]; [EMAIL PROTECTED]; Gregory Leblanc; bug1 Subject: RE: performance limitations of linux raid Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit There's "specs" and then there's real life. I have never seen a hard drive that could do this. I've got brand new IBM 7200rpm ATA66 drives and I can't seem to get them to do much better than 6-7mb/sec with either Win98, Win2000, or Linux. That's with Abit BH6, an Asus P3C2000, and Supermicro PIIIDME boards. And yes, I'm using an 80 conductor cable. I'm using Wintune on the windows platforms and bonnie on Linux to do benchmarks. I don't believe the specs either, because they are for the "ideal" case. Believe it. I was getting about 45MB/s writes and 14 MB/s reads using RAID0 with the 2.3.99pre kernels on a Dual PII 450 with two 30G DiamondMax (7200rpm Maxtor) ATA-66 drives connected to a Promise Ultra66 controller. Then I moved back to kernel 2.2.15-pre18 with the RAID and IDE patches and here are my results: RAID0 on Promise Card 2.2.15-pre18 (1200MB test) -- ---Sequential Output ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 6833 99.2 42532 44.4 18397 42.2 7227 98.3 47754 33.0 182.8 1.5 ** When doing _actual_ work (I/O bound reads on huge data sets), I often see sustained read performance as high as 50MB/s. Tests on the individual drives show 28+ MB/s. Sounds dang good, but I don't have any of those yet... When I can get a 1350 MHz proc, I'll grab a new machine an correlate these results for myself. :-) The performance is simply amazing -- even during real work (at least mine -- YMMV). And best of all, the whole set-up (Promise card + 2X Maxtor drives only cost me $550) I simply can't see how SCSI can compete with that. Easy, SCSI still competes. It's called redundancy and scalability. It's hard to get more than 4 (maybe 8 with a big system) IDE drives attached to one box. That same thing is trivial with SCSI, and you can even go with far more than that. Here's on example. At the office, I've got a single machine with 4 internal, hot-swap drives, and two external 5 disk chassis that are both full, as well as a tape drive, a CD-ROM, and a CD-RW. The tape is about 3 feet away, and the drive chassis are more like 12, everything is well withing spec for the SCSI on this machine. With IDE, I couldn't get that much space if I tried, and I wouldn't be likely to have the kind of online redundancy that I have with this machine. I'll admit that this is the biggest machine that we have, but we're only taking care of 250 people, with about a dozen people outside of the Information Services deparment who actually utilize the computing resources. Any remotely larger shop, or one with competent employees, could easily need server that scale well beyond this machine. I don't think that SCSI has a really good place on desktops, and it's use is limited when GOOD IDE is available for a workstation, but servers still have a demand for SCSI. Greg P.S. My employer probably wouldn't take kindly to those words, so I'm obviousally not representing them here.
Re: performance limitations of linux raid
I don't believe the specs either, because they are for the "ideal" case. However, I think that either your benchmark is flawed, or you've got a crappy controller. I have a (I think) 5400 RPM 4.5GB IBM SCA SCSI drive in a machine at home, and I can easily read at 7MB/sec from it under Solaris. Linux is slower, but that's because of the drivers for the SCSI controller. I haven't done any benchmarks on my IDE drives because I already know that they're SLOW. Greg Whatever you think of the interface (ide vs scsi) you have to accept that a drives speed is dependent on its rotation speed. A 7200RPM IDE drive is faster than a 5400RPM SCSI drive and a 1RPM SCSI drive is faster than a 7200RPM drive. If you have two 7200RPM drives, one scsi and one ide, each on there own channel, then they should be about the same speed. Multiple drives per channel give SCSI an edge purely because thats what the scsi bus was designed for. You pay a big dollars for this advantage though. Glenn McGrath
Re: performance limitations of linux raid
Clay Claiborne wrote: For what its worth, we recently built an 8 ide drive 280GB raid5 system. Benchmarking with HDBENCH we got 35.7MB/sec read and 29.87MB/sec write. With DBENCH and 1 client we got 44.5 MB/sec with 3 clients it dropped down to about 43MB/sec. The system is a 600Mhz P-3 on a ASUS P3C2000 with 256MB of ram, the raid drives are 40GB Maxtor DMA66, 7200 RPM, and each is run as master on its own channel. Turning on DMA seems to be the key. Benchmarking the individual drives with HDBENCH we got numbers like 2.57MB/sec read and 3.27MB/sec write with DMA off and it jumped up to 24.7MB/sec read and 24.2MB/sec write with it on. That, and enough processing power to see that paritity calc is not a bottleneck. Can you use your raid system with DMA turned on or do you get irq timouts like me ?
Re: performance limitations of linux raid
A 7200RPM IDE drive is faster than a 5400RPM SCSI drive and a 1RPM SCSI drive is faster than a 7200RPM drive. If you have two 7200RPM drives, one scsi and one ide, each on there own channel, then they should be about the same speed. Not entirely true - the DMA capabilities of IDE could provide faster transfer modes than your avg scsi card could generate. I have a 7200 RPM LVD scsi drive and a 7200RPM UDMA ide drive and the IDE wins EVERY SINGLE TIME. -sv
Re: performance limitations of linux raid
Seth Vidal wrote: A 7200RPM IDE drive is faster than a 5400RPM SCSI drive and a 1RPM SCSI drive is faster than a 7200RPM drive. If you have two 7200RPM drives, one scsi and one ide, each on there own channel, then they should be about the same speed. Not entirely true - the DMA capabilities of IDE could provide faster transfer modes than your avg scsi card could generate. I have a 7200 RPM LVD scsi drive and a 7200RPM UDMA ide drive and the IDE wins EVERY SINGLE TIME. -sv I think everyone seems to be missing the _point_ of SCSI. SCSI is NOT just about raw speed, even though some SCSI has a speed of 160MB/second. SCSI has many advantages over IDE, oe EIDE, such as command queing. Let's see your IDE drive handle 10 or 15 *simultaneous* reads/writes. let's see it have a MTBF of over 1 Million hours. Let's see it connected along with 13 other drives on the same chain. Let's see it run on cables running for several yards or meters. Let's see it conected to multiple computers. Heck, just for fun, let's run IP over IDE. Oh, wait, that's SCSI that can do that ;) The point is, comparing speed of SCSI vs any IDE variant is like comparing apples and oranges. That said, copmparing two drives of any variant, and basing their performance upon the rotational speed is also an error. RPMs are not the sole determining factor. Other factors include the size of the drive in MB/GB, and the seek or access time. I have, for example, a 5400RPM SCSI2 drive that still outperforms 7200RPM IDE drives colleagues have. I have seen 6X SCSI CDROMs that perform better than 8X EIDEs. I have a 7200RPM SCSI outperforming 10Krpm IDE drives. The determining factor in the choice of SCSI vs. IDE should be in what the machine will be doing. If all you need is a desktop machine fo rgaming and/or basic sufing/office type work, you probably don't need SCSI. if you are going to run a server that may be running some intensive I/O, and need reliability and longevity, you should be looking at SCSI, not IDE. After all, is speed is your sole criteria, chances are you are missing something. ;^) -- In flying I have learned that carelessness and overconfidence are usually far more dangerous than deliberately accepted risks. -- Wilbur Wright in a letter to his father, September 1900
Re: performance limitations of linux raid
bug1 wrote: I don't believe the specs either, because they are for the "ideal" case. However, I think that either your benchmark is flawed, or you've got a crappy controller. I have a (I think) 5400 RPM 4.5GB IBM SCA SCSI drive in a machine at home, and I can easily read at 7MB/sec from it under Solaris. Linux is slower, but that's because of the drivers for the SCSI controller. I haven't done any benchmarks on my IDE drives because I already know that they're SLOW. Greg Whatever you think of the interface (ide vs scsi) you have to accept that a drives speed is dependent on its rotation speed. Not entirely true. RPMs is but _one_ factor, not the determining factor. A 7200RPM IDE drive is faster than a 5400RPM SCSI drive and a 1RPM SCSI drive is faster than a 7200RPM drive. Not always true, if you are talking about data throughput and access speed. If you have two 7200RPM drives, one scsi and one ide, each on there own channel, then they should be about the same speed. Multiple drives per channel give SCSI an edge purely because thats what the scsi bus was designed for. You pay a big dollars for this advantage though. As well as various _other_ advantages, see otehr post. -- In flying I have learned that carelessness and overconfidence are usually far more dangerous than deliberately accepted risks. -- Wilbur Wright in a letter to his father, September 1900
Re: performance limitations of linux raid
Chris Mauritz wrote: Ive done some superficial performance tests using dd, 55MB/s write 12MB/s read, interestingly i did get 42MB/s write using just a 2 way ide raid0, and got 55MB/s write with one drive per channel on four channels (i had no problem writing, just reading) so surprisingly i dont think the drive interface is my bottleneck. I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Check http://www.tomshardware.com , they review the Promise IDE RAID card (the hacked one for $30). They get some pretty insane throughputs on some ATA66 drives. RAID0 seemed to scale rather linearly. I don't think there would be much of a problem getting over 100mbits/sec on an array of 8-10 ultra2 I tested a 3 drive RAID0 set where 1 disk = 1.7 M/sec, 2 disks = 3.4 M/sec, but 3 disks only = 3.9 M/sec. These were 5 M/sec disks on a AHA2940. I think SCSI bus overhead and contention issues play an important role. Ed -- Edward Schernau http://www.schernau.com Network Architect mailto:[EMAIL PROTECTED] Rational Computing Providence, RI, USA
Re: performance limitations of linux raid
Edward Schernau wrote: Chris Mauritz wrote: Ive done some superficial performance tests using dd, 55MB/s write 12MB/s read, interestingly i did get 42MB/s write using just a 2 way ide raid0, and got 55MB/s write with one drive per channel on four channels (i had no problem writing, just reading) so surprisingly i dont think the drive interface is my bottleneck. I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. Check http://www.tomshardware.com , they review the Promise IDE RAID card (the hacked one for $30). They get some pretty insane throughputs on some ATA66 drives. Gee, they look pretty awesome, excellent performance. The only drawback i can think of is that it wouldnt be as flexible as software raid. RAID0 seemed to scale rather linearly. I don't think there would be much of a problem getting over 100mbits/sec on an array of 8-10 ultra2 I tested a 3 drive RAID0 set where 1 disk = 1.7 M/sec, 2 disks = 3.4 M/sec, but 3 disks only = 3.9 M/sec. These were 5 M/sec disks on a AHA2940. I think SCSI bus overhead and contention issues play an important role. Ed I also see linear performance increas with two drives, but after that it tails off. Glenn
Re: performance limitations of linux raid
On Sun, 23 Apr 2000, bug1 wrote: Chris Mauritz wrote: Hi, im just wondering has anyone really explored the performance limitations of linux raid ? Ive just managed to setup a 4 way ide raid0 that works. The only way i can get it working is to use *two* drives per channel. I have to do this as i have concluded that i cannot use both my onboard hpt366 channels and my pci hpt366 channels together. Ive done some superficial performance tests using dd, 55MB/s write 12MB/s read, interestingly i did get 42MB/s write using just a 2 way ide raid0, and got 55MB/s write with one drive per channel on four channels (i had no problem writing, just reading) so surprisingly i dont think the drive interface is my bottleneck. IDE will always beat SCSI hands down for price/performance, but scsi is clearly the winner if you want 4 or more drives on one machine, as ide just doesnt scale. one of the ABIT boards comes with 4 onboard IDE controllers instead of the usual 2. As I recall, the kernel software will allow you to run up to 8 ide pci channels, and the 2.4 kernel reportedly will do more. Michael
performance limitations of linux raid
Hi, im just wondering has anyone really explored the performance limitations of linux raid ? Recognising ones limitations is the first step to overcomming them. Ive found that relative performance increases are better with less drives. Ive been using raid for a year or so, ive never managed to get a 4-way ide raid working efficiently (no timeouts), 2.3.99pre6pre5 is the best ive had though. Anyone have any thoughts on where the bottleneck is, or orther experiences with raid limitations ? Glenn McGrath
Re: performance limitations of linux raid
From [EMAIL PROTECTED] Sun Apr 23 01:06:24 2000 Chris Mauritz wrote: From [EMAIL PROTECTED] Sat Apr 22 21:37:37 2000 Hi, im just wondering has anyone really explored the performance limitations of linux raid ? Recognising ones limitations is the first step to overcomming them. Ive found that relative performance increases are better with less drives. Ive been using raid for a year or so, ive never managed to get a 4-way ide raid working efficiently (no timeouts), 2.3.99pre6pre5 is the best ive had though. Anyone have any thoughts on where the bottleneck is, or orther experiences with raid limitations ? I've not had any real problems using striped sets of SCSI drives as RAID 0 and RAID 5. You're always going to get rather crappy performance with lots of IDE drives unless you have only 1 drive per channel. By the time you buy that many controllers, the cost is pretty much a wash with SCSI. Ive just managed to setup a 4 way ide raid0 that works. The only way i can get it working is to use *two* drives per channel. I have to do this as i have concluded that i cannot use both my onboard hpt366 channels and my pci hpt366 channels together. Ive done some superficial performance tests using dd, 55MB/s write 12MB/s read, interestingly i did get 42MB/s write using just a 2 way ide raid0, and got 55MB/s write with one drive per channel on four channels (i had no problem writing, just reading) so surprisingly i dont think the drive interface is my bottleneck. I find those numbers rather hard to believe. I've not yet heard of a disk (IDE or SCSI) that can reliably dump 22mb/sec which is what your 2 drive setup implies. Something isn't right. I think read performance is a known problem, but at least i dont get lockups or timeouts anymore. IDE will always beat SCSI hands down for price/performance, but scsi is clearly the winner if you want 4 or more drives on one machine, as ide just doesnt scale. It's not so hands down anymore. SCSI drives are becoming quite cheap. Yes, IDE is cheaper, but not THAT much cheaper...especially if you want to use more than a couple of disks. SCSI drives are 50% dearer than ide arent they ? What sort of performance do you get from your scsi sets ? I was getting 45-50mb/sec with a striped set of 3 50gig Seagate barracudas using software RAID0 and an Adaptec 2940U2W controller. I used bonnie to test and used 4-5 times the system memory for the data size. I wonder what the fastest speed any linux software raid has gotten, it would be great if the limitation was a hardware limitation i.e. cpu, (scsi/ide) interface speed, number of (scsi/ide) interfaces, drive speed. It would be interesting to see how close software raid could get to its hardware limitations. RAID0 seemed to scale rather linearly. I don't think there would be much of a problem getting over 100mbits/sec on an array of 8-10 ultra2 wide drives. I ultimately stopped fiddling with software RAID on my production boxes as I needed something that would reliably do hot swapping of dead drives. So I've switched to using Mylex ExtremeRAID 1100 cards instead (certainly not the card you want to use for low budget applications...heh). Cheers, Chris -- Christopher Mauritz [EMAIL PROTECTED]