Date: Wed, 22 Mar 1995 11:15:44 -0800 (PST) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: sos@login.dknet.dk (S|ren Schmidt) Cc: hackers@FreeBSD.org Subject: Re: Why IDE is bad Message-ID: <199503221915.LAA09736@gndrsh.aac.dev.com> In-Reply-To: <9503221217.AA08636@login.dknet.dk> from "S|ren Schmidt" at Mar 22, 95 01:17:17 pm
next in thread | previous in thread | raw e-mail | index | archive | help
[cc: trimmed, moved to hackers list, reply-to set to hackers] > > > > > I don't think you will find many ide's faster than the WDC, I know > > it can run faster on a better IDE-controller, but this is a fairly > > standard IDE controller so it's a good indication. > > And yes, but FreeBSD doesn't support DMA in IDE (yet ?)... > > Was that my que word ?? > > DMA does not pay on E-IDE devices, it's not busmaster DMA, it > is the motherboards DMA we are talking about here. Yep. > This is very slow on standard MB's, but faster on PCI > (if they implemented it right). Yep. > However a "generic" driver would have to fall back to > the std. PC DMA and that is slower than using polled mode. Perhaps, but motherboard DMA does occur at upto 4MB/sec, and concurrent with the CPU doing other things. Since most drives can't sustain this transfer rate for long periods of time it would be more efficient. > And besides only few drives support it anyway, and the high > rates of the PIO modes looks better in advertizing. E-IDE DMA mode is not a function of the drive, it is a function of the controller chip. > And that is the main thing on E-IDE, the drives are designed > with enough onboard cache, that coretest etc. reports transfer > rates close to the interface speed (13MB sec or so), but the > drive cannot hold this speed when it has to read from the media. All too true, PC hardware is optimized to run the Benchmarks that you see in all the trade rags, not for real world applications. Being fast on the benchmarks does often help some, but it is no where near what you can get out of good hardware by spending the extra money to get what is not only fast on the dumb benchmarks, but is fast no matter what you thow at it. > And here is the catch, in that most el cheapo IDE drives has > inferior drive mechanics (hey they are cheap), and then some > fancy cache/interface electronics to make up for outdated > hardware.... I don't think this is totally true, many very good drives are avaliable in both SCSI and IDE, they use the exact same technology just different drive electronics. Same samples are the Micropollis 4410, the Quantum ProDrive, Lightning and Maverik series. I won't call Samsung a good drive, but they also have a line of drives that are avaliable in either SCSI or EIDE. > > > I personally think that E_IDE and friends are the cleverest > sales trick in todays PC hardware. BINGO!!! And the other problems with E-IDE or even IDE is that it is hard to get the real numbers you need to know if the drive is really fast or not. Drive adds always list the host bus burst interface rate, what you REALLY want to know is the ``Data Transfer Rate - To/From Media''. This number is almost always avaliable in the specs for SCSI drives, and only sometimes avaliable in the specs for ATA drives. I happen to have the numbers for the Samsung Transfer Rate drive ATA/SCSI Host(SCSI/ATA)/sec Media TO/FROM /sec SHD-30280A ATA only X/11MB 38.10Mbit SHD-30420A ATA only X/11MB 38.10Mbit SHD-30560A ATA only X/11MB 38.10MBit PLS-30540A ATA only X/11MB 46.22MBit PLS-30730A ATA only X/11MB 46.64MBit PLS-30850A/S ATA/SCSI 11MB/10MB 48.11MBIT PLS-31100A/S ATA/SCSI 11MB/10MB 47.27MBIT TRS-31080A/S ATA/SCSI 11MB/10MB 38.5MBIT/72.4MBIT Here is a line of drives that should do sustained tranfers of 4MB to 9MB/sec avaliable in both SCSI and ATA. Me personally, only use ATA when I have to and/or when the customer is cutting cost. I wonder what Poul's little test would like like if I compared the two versions of that TRS-31080, how much CPU would I need to keep up with 9MB/sec PIO :-) -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Custom computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199503221915.LAA09736>