Slashdot Mirror
Will Firewire be the death of SCSI?
cholko writes "Over on
TechWeb there is a story about the use of Firewire in the new Macs. What makes the story interesting is the comment about the possibility of SCSI being dead within 2 years.
Firewire will eventually supplant USB as well, but apparently IDE will be around a while in the mass storage market."
Just like usb was gonna replace serial? How long has this been out?
the only wire on fire is in my pants
SCSI dead my ass. I can see as
firewire grows the price on SCSI
will drop in order to keep up
competition. Looking forward to
10GB Ultra SCSI 3 HD for $300.
Maybe Adaptec will get a clue and
drop the price on their controller
cards. Cant be more then $30-$40
in parts on those cards. Giving
them a *nice* healthy profit margin.
Just how much cheaper has SCSI become over the past few months? or years? It seems to me that SCSI will never become mainstream... It's too costly to produce.. :)
I'm just looking forward to having a K7 with USB
Why does every new technology have to kill off some older technology? Firewire sounds cool. It isn't as touchy as SCSI, and the cable is easier to work with, making it great for things like (surprise) digital cameras. But that doesn't mean SCSI will suddenly stop working, or lose performance, or have OEM support evaporate. Radio didn't disappear when TV was invented, either.
I never thought Firewire was supposed to compete with USB, aren't they rather complimentary technologies? You throw your video devices and hard drives on firewire, you put your mouse and joysticks on USB.
Next, isn't Firewire (and USB) part of the Device Bay standard. Is that dead already or will it actually show it's head soon?
And the biggest obstacle I see to Firewire replacing SCSI is that people who have SCSI systems don't often have just a small HD hooked up to that system. Good thing I have a wide controller so I can use those other 8 devices, as I already have 7 SCSI devices in there. As much as I'd like to try the Firewire thing out, it'd cost me a fortune to switch. I can see myself buying new devices with Firewire connections though. But what do I need other than more HD space?
It does have a shot with the device bay though. That or if it can be added to mobo's for essentially free (ala USB) and a lot of people have it without realizing it. Then bring out the drives for the same cost as SCSI. Or probably a little more cause external would propbably be the main choice.
As a side note, I've seen a recent Compaq with Firewire connectors on it. Supposedly, they don't yet have an adapter available so that you can actually hook up any devices.
Firewire replacing USB would be as ridiculous as SCSI replacing RS232 - the two are meant for completely different markets.
Why on earth would I want to hook up a mouse, keyboard, speaker, etc to a 240 Mb/s connection???
That's part of the point... one connector for
all devices. Simplification is the goal.
$15,000 worth of SCSI peripherals in my graphics workshop is a good reason...
Adaptec get a clue - don't think so.
They killed all firewire development about
a year ago, aka IEEE 1394. They have gotten
rid of the majority of the R&D effort in
this last year because of the significant
drop in prices. They have gone from a billion
dollar company this time last year to perhaps
a 400 million next year - and I suspect
even further reversals in their future.
(Strip away R&D...no new products..)
They are now exclusively focused on SCSI and
building high-end RAID hardware and have no
presence in the average desk-top market. Who
is going to pay $160 for a SCSI controller
when the motheboard comes with a 66Mbyte/sec
EIDE controller built in?
IDE is getting faster - 1394 will be built
into the MB too...and SCSI will wither in
2 years - that time frame sounds just about
right.
So what? You're still wrong...
Anyone building a high performance system will definately be inclined to use FireWire if it's the same cost as SCSI. The speed is plenty fast enough for within a rack of drives (8 or so), and the cable is *MUCH* easier to deal with. SCSI cable is an enormous unbelievable pain in the ass.
Now, FireWire isn't fast enough for connecting drive racks to controllers, so traditionally SCSI would have to be used for this purpose. All else being equal SCSI becomes more convenient. All else isn't equal because fibre channel is currently putting everything on its head. Now we have firewire in the racks and fibre channel outside. No more touchy cables, longer cable lengths (*MUCH* longer it turns out) and higher bandwidth to boot! Bye bye SCSI...
Compared to fibre channel speeds, SCSI is like my grandmother in a 100m dash. And my grandmother is dead. If FireWire catches on SCSI is going to get squeezed out of the middle.
And exactly how fast do you piss man?! Seems to me a good drinking straw would be fine...
These days IDE and SCSI drives rarely use the same mechanics - though they did for a while. The SCSI drives are now all 7200rpm or 10000rpm, with the fastest head actuators. The IDE drives are mostly 5400rpm, with slower head drives. The SCSI drives usually have larger buffers. The price differential is MUCH less than 300%. All that said, the SCSI drives should be cheaper, and would be if they were made in the same volume as IDE.
IDE systems can, and often do, use a slightly modified form of SCSI interface between drive and host. The commands are the same, but the data transfering mechanism is different. The key performance difference in multi-tasking system comes from IDE's one by one approach to processing commands.
Firwire runs a 100, 200 or 400mbps. SCSI currently runs at 1280mbps, and uses its bus more efficiently. Firewire is life in the slow lane. It also has a dumb connector that isn't too robust on a minDV HandyCam.
Some technologies reach natural limits.
For SCSI - I'm betting that they've
used about all the tricks they can
without loosing ALL backwards compatibility.
SCSI-160 or whatever it's going to be
called is using the LVD driver technology
AND using both edges of the REQ signal
for moving data. They're going to have
to change the signaling system RADICALLY
to go beyond these tricks to move data
faster.
While at the same time - Firewire will be
at 800Mb/s fairly quickly - but it's
tough to do that... at least for ASIC
houses with nominal experience. It's going
to take some proprietary IP from the
experts to go beyond that... thus get
real expensive.
I think SCSI has breathed it's last gasp
High end servers aren't conservative
at all - they tend to use the most
blazingly fast stuff they can find.
Top performing RAID servers are using
Fiberchannel today! If SCSI in it's
160Mb/s aspect surpasses Fiber for the
time being - they'll switch as soon as
a drive manufacturer bellies up to the
bar.
Be careful! That's the speed of the BUS, not the DRIVE. A typical high end Seagate or IBM HD might seek for 6.3ms avg. and spin at 10,000_ (maybe 14k+ now?) rpm. That's the actual limit on SCSI hard drives. A good drive might be able to sustain 7-10 megs/sec if it almost never has to seek (100k continous blocks or so). Note that if you read 10k blocks you only get 1.5 megs/sec.
OK, if that's true, if you raid 4 drives(even if they're not raided, lets just say you access 4 drives at the same time), you'll tie up that 40 meg of bandwidth on the controller. 10 meg a second sustained throughput is not at all rediculous, I have an ide drive that can sustain that for sequential access without a problem. And that's just a cheapie 200 dollar 8 gig.
How much cpu usage is involved in firewire? I know ide chews upwards of 4-5 times as much cpu time as scsi for similar throughput(at least on my hardware). For me at least(and I would imagine any place where they would use scsi), cpu usage is probably the most important aspect of scsi.
Anywho, what was the limit on firewire's throughput? 50 meg or something I believe someone said? So that limits you to 5 10 meg/sec drives before you start to hit a performace cap. I don't think we have many servers here that have less than 8 ultra2 scsi drives, and I'm sure they all pull 10 meg a second. So How is firewire going to replace those?
Just a few cents...
I'd think of putting in another SCSI card and
splitting the load if I were you, assuming you have a spare slot... Just because you _can_ run
multiple devices on a single host adapter, doesn't mean you _should_..
I think the point is that in 5 years your $15,000 dollars worth of equipment will be swapmeat and Firewire (the higher grade stuff, 1600+) will be the dominant high end bus with pretty much all new periphials coming out for it.
Question: do you still have to deal with drive IDs in FCAL? IIRC SSA doesn't have this limitation..
Of course, I doubt SSA is ever going to be faster than 80MB/s...
Er, you got IDE vs SCSI backwards: SCSI came first, and IDE started its attack on the low end..
Where to get it? Please.
Fibre Channel is FireWire, sort of like how Mac calls PCMCIA cards PC cards. Its part of the device bay standard and I heard the Windows 98 SP is going to add more support for it. And its not meant to replace USB. And what else, oh Adeptc does have a FireWire card, look at their product line here
> I'd worry more about seek and spindle than Ultra3...
My Man! Data/Index/Metadata placement on separate spindles/access arms! Be still my beating heart!
How many years have I worked with hardware weenies who don't understand why SQL and RAID5 don't mix.
All they know is physical and logical partitions.
Transfer rates are for videos downloaded from alt.pictures.badstuff.
63 separate devices are for serious databases.
- No you don't need two cards to get 80MBps SCSI.
- Who cares if it supports 63 devices, lets say you are using el-cheapo 5400rpm WD drives with a maximum sustainable transfer rate of 5MBps now on 400Mbps firewire that gives you enough bandwidth for 10 drives (keep in mind those are cheap drives, with nice Quantum drives I wouldn't put more than 4-6 on a controller)
- Yes SCSI is (to some degree) hot-swappable. The people running systems high-end enough that they care about bus throughput are probably running RAIDs and they most certainly didn't design them so that they have to take down the storage array to swap a drive.
- You can hook a SCSI system up to as many computers as you want. It's called NFS. (or if you must use inferior protocols SMB or EtherTalk)
Yeah Firewire is great and I'm sure I'll get it on my next 'puter. It would work well for the video, audio, and high-bandwith periphrials, but the day Sun replaces the SCSI arrays in the UltraEnterprise 10,000 Starfire storage arrays with firewire will be right after the G3 processor is "The fastest processor available."in fact FireWire is so good that after Apple
invented it (how ever many years ago) they
immediately put it on all their systems.
and when they did this wonderful new technology on the motherboard, they decided to try and kick
start some serious availability by using it
for all the internal drives.
_NOT_
I admit that hot plug is nice (although most
SCA drives are hot-pluggable now) but despite what some journalists appear to think (mostly they dont appear to know their bits from their bytes!) scsi is faster. and i can order a scsi disk drive from one of the major manufacturers and have it delivered tomorrow.
Apple are probably making a tactical mistake - again - by not for example speccing the systems with Firewire primary hard drives (keep the IDE interface on the m/b, sure, but why use that crappy technology when you have something much better) which would provide a quick boost to availability rather than customers having to wait for things to trickle out over the next several months to years...
Whoa!! Installed base is prolly the singled biggest factor in this industry, with the possible exception of marketing. Why else would banks still be using the same mainframe computer systems they have since 1960? Why else was ompatability with a glorrified interrupt handler from 1978 (DOS) be considered vital until two years ago? Why else is everyone *still* using 1.44 MB floppy disks?
Advantages/disadvantages to SCSI and firewire aside, anyone who ignores their installed base will find their career short-lived.
SCSI will have to get lean and mean if it wants to hold on. It's a great technology and there are a lot of scsi products but it's the new devices that get made that will kill it. SCSI has enjoyed a kind of monopoly position, it is over priced because they can get away with it. With standard firewire on every new computer instead of IDE SCSI will have to start offering something really compelling or become dirt cheap if it is to survive. As wonderful as it is, SCSI is still pretty confusing from a driver perspective, how many different chipsets do you need to support before you can say you support scsi? With firewire there will probably be one firewire driver.
Future computers will look like they should have looked already: USB driving keyboard, mouse, printer, zip drive, joystick, scanner, etc..
Firewire running the drive stack and other high bandwidth devices. No other plugs cluttering or confusing things.
> run IP over fibre channel
And you can run IP over SCSI. Don't have the URL handy, but somebody did this to provide a faster network to interconnect a cluster.
As other posters pointed out, FireWire is not a competitor to USB but rather a complimentary technology. (no one would want to hook up a mouse to 400 MB /s port...)
As for FireWire taking on SCSI... I don't think it's gonna win any time soon. SCSI is faster and IMHO it will remain the interface of choice for high-end servers. (Of course one might argue that Fibre Channel will be the king of the REAL high end...)
What I'd rather see happening (and I'm pretty sure it will) is FireWire replacing IDE (lewt this beast die!). It may happen soon because from what I understand you can hook up an IDE drive to FireWire. This would be really great because as of now, the fastest external port PCs have is the parallel port (yukk!) and you can rarely (never?) find SCSI on home PCs. If FireWire were to become a standard, it would really great for external storage media, etc.
People use SCSI for more than just hard drives (although you'd never know it from reading most of the above comments). Modems, video equipment, sound equipment, scanners, etc, all use SCSI. For these things, FireWire is and EXCELLENT replacement for SCSI, mostly because it's hot-swappable, and you don't have to diddle with ID's and other SCSI silliness. Long cable lengths and other things are also nice. Hard drives puking at 40 MB/second ain't that bad either - especially for 99.9% of the population.
The only place where SCSI may continue to thrive is on the high-end server/workstation, where power is everything.
Why FireWire is Hot Hot Hot:
http://www.vxm.com/21R.35.html
This has gotten a bit out of hand. SCSI will take eons to do die, and -if- it dies, rest assured it won't be because of your pissy little FireWire (herein abbreviated to FW).
SCSI is a *lot* faster, more evolved, has devices actually MADE for it, supports internal/external devices bloody brilliantly, and is an all-round Damn Good Thing (tm).
IDE is a *lot* slower, less evolved, has devices which use *your* CPU time, doesn't support external devices (if it's parallel port and it's not a printer, it should piss off), slows down the bus as you add more devices, can only handle 4 devices anyway (no, those Promise cards with more IDE channels don't count), and in generally sucks ass. And seeing as PC99 is written primarily by Intel and Microsoft, I don't really see how people can think it's superior -- it's just what they want you to buy. Apart from the pricing scheme, there's no argument as to use IDE over SCSI. Incidentally, it's just the manufacturers being assholes about the pricing, there's no technical reason for the prices being so inflated.
Fibre Channel rocks. It's bloody fast, already in use (as previous people have stated), and supports a lot more features than FireWire ever will. I mean, do you *really* have 127 devices that you want to plugin that aren't HDD's? I sure don't.
So no, I don't see the death of SCSI happening any time soon. And as for USB, well, I see that becoming a standard at about the same time they a) kill the serial/parallel buses b) use only LCD monitors in new machines c) kill ISA d) make PC's with more than 1 AGP slot and d) hell freezes over. Get the picture?
Some factoids from FC-AL land:
:).
Real application level data rate which I was
able to achieve is 92MB/s (close to promised
100, but not quite
UltraSCSI has data rate of 40MB/s but doubt
it works this way. On regular fast/wide we
get about 17 out of rated 20.
Next development in SCSI is a low voltage
SCSI with higher clock rate, it approaches
FibreChannel: 80MB/s rated.
--P
Additionally, IDE is only capable os single threaded IO, versus SCSI's multithreaded IO. Also IDE devices can't negotiate the bus in an efficient manner. Sorry, just had to throw in my 2 cents. I am not sure premis of the notion that 1394 will replace SCSI is sound. It will still take a year or two for 1394 to become prevelant (assuming the article is correct). During this time it is possible for SCSI to be enhanced to overcome it's big hurdles compared to 1394, cost, number of devices, and cable length. Cost can be reduced through greater integration (it is probably already there, the profit margins are artificially high, I think). They need to work on cable length, which would seem to forfit backward compatibility (at least if you mix devices). The latter would also apply to number of devices. Anyway, the SCSI standard is not static, that was my point.
At current, the device bus landscape looks
like this
Serial:
slow, crap, can barely keep up with the
modem on the other end -- ok for mice, but
domb -- very dumb.
Parallel:
dumb by design, and being hacked into the
ground just to get the flexibility of SCSI-1
IDE:
cheap and cheerful
SCSI:
better
And the new interfaces stack up like so...
USB: Set to replace
Serial, Paralled, slow uses of IDE (like
ZIP's) etc
Firewire: Set to replace
IDE (for hd's and cd's)
Parallel and ethernet and SCSI local printers
SCSI for HD's (with the notable exception
of Fibre Channel SCSI)
Current FireWire is 1394a. This specifies speeds of 100, 200, and 400 Mbits/sec. The next spec is 1394b, which is 800, 1600, and 3200 Mbits/sec.
I'm not sure about the time frame af any of the 1394b stuff. About a year ago I was in a project that had a 1600Mbit/sec reciever and transmitter untill Intel pulled the plug. I think TI might have something ready, just not in production.
In two years - how do you want to be hooking up your drives? With big SCSI cables; having to power-down to plug it in; termination problems; ID conflicts -- or a single plug for fast data, hot-plugable, and configures itself? Technology should make life simpler/easier - Firewire does that compared to SCSI. Those who like to fiddle with IDs and terminators can keep them - I'll take Firewire thank you.
FireWire can build on the existing SCSI infrastructure (by encapsulating SCSI protocols), but should make the kinds of devices that SCSI is used for now much more usable and accessible.
For the really high end, FibreChannel is likely to be the winner.
Actually addressing extensions in FC-AL are
:)
just too big for perfect compatibility.
For example even on first level FC-AL
array gets 14 bit LUN number, or 16 millions
volumes. Sun's own A7000 runs out of LUNs
with Fibre Channel headend (there is a project
to fix it underway).
Number of targets is limited to 126 in case
of FC-AL but for fabrics you have 24 bits.
That is a lot of targets. Conventional systems
cannot handle this properly. Either switch
must map addresses for them, or host software
needs to be converted more into network-like
software, or both.
Also as one FW proponent noticed above people
want to connect many hosts to many peripherals.
Dunno if Apple programmers are geniuses
(perhaps they are), but Sun has difficulties
with this concept applied to conventional
peripherals. Just how many commands can you
queue in a single drive? Each Seagate drive
has a hard limit and when several hosts talk
to the same drive they overwhelm it. Essentially
drive becomes a server and we know that servers
are subtle business (for exmaple Slashdot
crashes
--P
The point is to run IP over SCSI over FC-AL.
Unfortunately the author of current IP-on-SCSI
in Linux smokes weed and fails to understand
that.
IP over FC-AL requires all adapters to handle
incoming long frames with some sort of queue.
Ethernet boards all do this of course, but FC
boards do not. This way FC boards may be
"IP Capable" or not. Still the speed is there
you only need to unleash it with IP over SCSI.
In the long run IP over SCSI is a dump idea
because of high overhead. But it may work for
some time.
--P
RAID5 and SQL mix perfecly in _modern_
databases. One thing which you need to
put aside is database log. And who keeps
database log on a mechanical disk anyways?
When a database commits a log it writes
full stripes into RAID-5 with a speed of
any other RAID.
I tell you _very_ many people buy Sun A5000
and place Oracle tablespace into RAID-5.
It works just fine as long as you move
hot spots out into a solid state device.
--P
Firewire will not take the place of SCSI in the high performance market - Fibre Channel will. In fact, in the world of really high performance - disk arrays with dozens of ports, multi-TB capacities, multi-GB cache memories, this is already happening. The emergence of System Area Network technologies (still in their infancy, but some of us are working on that) is also driving the adoption of more topologically flexible technologies, such as FC and FW but not SCSI and absolutely not IDE.
That's the problem with Firewire. Between IDE at the low end, FC at the high end, and SCSI already ensconced in between, there's no "turf" that's not already staked out. It's a shame, too, because technically Firewire is the most elegant of the bunch (followed by FC, then SCSI, then IDE as the grungiest and least general of the bunch). SSA also scored a few technical merit points before it died, for incorporating many of the same ideas as FW and FC.
Wow finaly I hear a voice of reason on the Slashdot... Yes but of course FC is not
consumer friendly. This is what we see today:
FC-AL drivers are being sold alongside with
Sony cameras with 1394. And I think there is
nothing wrong with that. SCSI is going to
die out because LVD came a bit too late.
--P
Just about every modern disk technology except IDE already uses a SCSI command set regardless of the transport. Why? Elementary, my dear Watson. Manufacturers don't want to manufacture drives to a bunch of specs that are different from top to bottom. They want the command-interpretation part of the controller to be the same no matter what, then slap on whatever flavor of glue they need for the transport du jour. And of course the disk manufacturers are well represented on the standards committees, so the standards accomodate this desire rather well.
We did this at Clam Associates for HACMP/6000, and I'm sure we weren't the only ones. It wasn't for higher speed, though (see below), it was just to make use of every possible channel that we could to prevent the problems associated with the cluster becoming totally partitioned.
The problem with IP over SCSI is that, while througput's fine, latency bites. For request/response small-message traffic you are quite seriously going to see better numbers on a plain old 10Mb/s Ethernet.
The Mb/s throughput numbers for SCSI may look fine, but the usec latency numbers look pretty awful.
Unfortunately for us optics comes to higher
speeds first, then copper pulls up there.
There are many reasons for that, mostly
difficulties with crosstalk, noise, and EMI
in copper environment.
FC has copper defined and working, but it is
not the kind of copper we all are used to.
It runs over a really expensive STP cable
which is not any better than 62.5 um fiber.
Ethernet has optic PMD for Gigabit which they
borrowed from FC. This actually means that
Ethernet may use FC copper. But IEEE is reluctant
to bless this because there is no advantage
to it.
BTW, FC defines a geekish piece of equipment,
a GBIC. It is a tranciever alike AUI-to-10baseT.
As all of you know nobody use transievers today.
100base-T comes right off a motherboard because
it is cheaper. Still some peolle use GBIC.
I personally hotswapped them between a Sun
A5000 array (FC) and Extreme Networks switch
(Ethernet). That is a level of compatibility
which I like! Too bad GBICs are expensive.
And for the record: it is a pity that folks
think that FW and FC are the same thing.
--P
I believe the following to be the case:
All SCSI command phases run at the speed of the lowest bus on the device, so that if you're running transaction-intensive operations (as opposed to transfer-intensive), you'll see the hit of that slow device more often than you'd like.
Ummm, yes thats right check your stats & math.
400 Mbs = 50 MBs [Now]
1600 Mbs = 200 MBs [Later]
Gumber sez:
IDE was cheap, it required little supporting circitry on the disk or the motherboard.
SCSI cost more.
Cost is an important factor in engineering.
I have seen SCSI sound gear... samplers and other pro audio equipment typically hook to the scsi port.
Although I don't know what the hell he was talking about with regards to the modems... a SCSI modem? Don't think so, pal.
Congradulations, that is the longest damn comment i've ever seen
:Oh, you must be stuck in the low end consumer market, while I'm stuck in the high end market
My hard drive is bigger than your hard drive... nanner nanner nanner.
It's just the geeks way of measuring their dicks.
See above
I can sure see FireWire replacing external SCSI, but internal SCSI will still be around. FireWire won't be replacing internal SCSI anytime soon.
If anything, FireWire (aka IEEE 1394) will quickly replace things like Fiber Channel and the other wacko high-end drive mechanisms. Since Adaptec is one of the driving forces behind FW, I don't think that they want to see their bread-and-butter (SCSI cards and chips) go away anytime soon.
The reason that USB isn't as big a hit as it should be is that only one PC OS has good support for it - Win98. Oh yea. That works. As part of some testing, I installed a pair of Labtec USB speakers in a Win98 system (no other sound card) and it worked perfectly, from the typical windows sounds to handling CD-Audio (it does CDDA from the CD-ROM to the speakers).
So why doesn't Linux have USB support? Easy. Few USB devices to begin with, and nothing to compare the usage with. Given that Adaptec has opened up, and I'm sure Apple wants us Linux people to buy their stuff, you'll probably see more FW support coming out for Linux.
Huh? FireWire is supposed to hit 1.2GB soon. Last I heard (this was about 18 mos ago) the silicon was being made for 800Mhz.
Lets see now, scsi is up to 40 (80? 160?) megaBYTES a second. 40*8=240megabits. looks like 80megabyte scsi beats firewire already. Then we remember that 80megabyte scsi is still compatable with scsi-1 at 5 megabytes, and will be compatable with all faster scsi.
I expect scsi to die, but it won't be because of firewire so much as because of fibre channel. (normally 1 gegaBIT, but up to 4 gegabits) And then it will only die because electrical engineers hate dealing with the fact that at these speeds a signal will arrive at the end of one wire faster then at the end of the next.
I can't wait for IDE to die. It never should have been invented, scsi was always better. Okay, once in a while ATA/IDE had a faster therotial transfer rate, but scsi could come closer to its theoritical maximun.
Accually I do expect fibre channel to replace scsi. It already is, because all the big venders are commited to it. BTW, by big venders I don't mean the seagate's of the world. I mean the big RAID people, who would rather sell fibre channel then anything else. The server venders are coming around to because the fibre channel message has been preached enough.
You can of course by most fibre channel products in scsi versions, and often SSA (IBM's thing) or escon (IBM again, but a little more open) but the venders want to move it all to fibre channel.
Need I remind everyone that scsi is normally used in the big high end servers. Scsi is more expensive even when the parts are IDE with a different board because the customer is willing to pay for the feeling that it is more reliable. (lets not get into the scsi-ide arguement, but for these people scsi does have an advantage above reliability which may or may not exist)
Note that I'm talking five years out though. Scsi will be sold for a long time yet, the best case perdictions say scsi will be around in the high end buiness for at least five years. I belive they are too optimistic, but that is why they are best case perdictions.
I have no idea what firewire is, but I do know that there is no interest in the high end.
BTW, in interest of disclosure, I will benifit if fibre channel makes it big. I also personally belive in fibre channel.
Huh? Who still uses that old 8 bit scsi stuff? Oh, you must be stuck in the low end consumer market, while I'm stuck in the high end market.
Source: An in-depth introduction to scsi by David Demming. (I don't ahve my copy here, but I took a class tought by the author who is on the scsi committie) 80 megabytes/sec was achived in the real world over a year ago.
And I'm smarting for my math mistake up there, that someone pointed out... :(
Go over 6 BUSY hosts, and ethernet drops to about 30mbits/sec due to colisions.
Your switch arguements is irrelavent. Ethernet has nothing to do with switches, I could build a scsi switch if you gave me the money. It wouldn't make sense (due to the 16 device limit and the rarity of putting multipul hosts on scsi, but it can be done) Most computers are cabiable of acting as a router. Put a few scsi cards in several comptuers, and you can route ip. Hint, IP doesn't care what it is going over.
Come to think of it, ethernet is a bus just like scsi. (except slower) Note that I agree that ip over scsi doesn't have much a point.
How long were CD-Rs out before they became popular? Or Minidisk? Or 3.5" floppies? Or mice for that matter.
Just because it has been out for a while does not make it vaporware. It takes a while for somethings to take off...and others like DVD and DSS dishes take off like a rocket.
Firewire is going to be the next standard connection for home theater too.
The new G3 Power Macs will go to the 400 Mbit/sec.
And they don't require extra hardware. For some of those thumpin' throughput SCSI cards don't you need two of the cards? Are they hot-plugable? I don't think so. Do they support 63 devices? I don't think so.
Now I think I remeber seeing them share a camcorder with two computers and a hard disk with two computers at MacWorld...I know you can't do that with SCSI.
Anymore than Fibrechannel replaced SCSI. SCSI is a protocol as much as an electrical specification. What you will see is Firewire being the communications protocol SCSI is run over (except for the high end, where FibreChannel and Storage Area Networks will still rule).
Firewire is not going to replace USB. USB is cheaper than Firewire by quite a bit. For low-bandwidth devices (like keyboards and mice) it's not worth it- save the $20.
Firewire may replace PCI. I don't think so, but it may. The next revision of Firewire will be spec'd to 1.6Gbit/sec ~= 160MByte/sec for 33MHz 32bit PCI. PCI, however, has already been Spec'ed up to 64 bit and 66Mhz, giving 400MB/sec performance. What I'm betting will happen is that 64bit 66MHz PCI will become the standard for "high speed, no hotswap" peripherals like video, and for tying together multiple Firewire buses. Firewire (running SCSI among other things) will become the standard for "medium speed, hotswappable" peripherals like hard disks. USB will become the standard for "low speed, ultra low cost" peripherals like keyboards and mice.
All the SCSI standards nowadays are annoying as shit. Not to mention termination, adapters, mixed cables and connectors... Bah. Good riddance. Of course I love SCSI over IDE - but I'll take some FireWire.
Not like a keyboard, speaker, mouse or even a modem would even put a dent that bandwidth.
Yeah, exactly... Hot plug and play is nice. RAID I think would still be a manufacturer option though, no?
IDE's future is (unfortunately) rosy: PC 99 (I think, might be prior to that) requires an IDE interface on the computer EVEN IF it's got a SCSI one. This basically means most computers will just have IDE (sub-$1000 trend: minimize extras).
Check out the ZD coverage of Steves address, very interesting stuff. Steve explains it well.
USB 12 Mbits per sec 127 devices
Firewire 400 Mbits per sec 63 devices (COOL!!!!)
The Mac's has two of each ! damm
This looks great to me. If we let Intel dole out technology in small increments.
Linux is only free if your time has no value. Windows is only free if you threaten to use Linux.
My Seagate 18gb 7200 rpm disks move 15mb/sec to/from the media. 3 of these drives running at full throttle can completely soak out my measily 40mB/sec (320mbit/sec) ultrawide SCSI. Even if 400mbit/sec firewire was available right now, it still wouldn't be much of an improvement.
Firewire's going to be great for scanners, digital cameras, and the like, but if you have a big stack of large capacity 10,000rpm disks connected to your server, you want ultra2wide SCSI 80mB/sec (640mbit/sec), not 400mbit/sec firewire.
People assuming USB will be absorbed by Firewire are missing a few
points about the whole purpose of USB. Its designed to be a cheap and
flexible connector for low bandwidth devices, most especially devices
that aren't expected to have an increase in bandwidth demands over
time, and eliminate all other connectors in that area.
Keyboards have not significantly expanded their bandwidth demands over
time. Likewise mice, game controllers and so on. You might make a case
for speakers, microphones and suchlike, but even then there's a law of
diminishing returns.
Firewire is aimed at the category of devices that like to use all the
bandwidth available, and show benefits from said approach, like hard
drives, video feed equipment and such.
Sure, you could stick a keyboard on a Firewire line but what's the
point? The connector is likely to be more expensive and now you have
to worry about your hard disk slowing down your typing speed. And
given that the demands of technology will eventually force someone to
design a whole new connector better than Firewire, you have to go and
redesign your keyboards as well.
Connectors last until there's a real need to replace them. If USB is
good enough as a connector, it can last next to forever. People who
manufacture USB devices won't move off USB until its clear that their
devices are going to benefit from it or they can't continue as they
have in the past.
USB is getting established because there are limitations to serial,
parallel and a list of other ports. USB can do things they can't and
its a single standard they can all adhere to. Manufacturers and
consumers will like having a single ubiquitous standard and won't move
off of it until they have to.
In the case of Firewire, people are likely to keep demanding bigger
pipes for all those high-bandwidth devices they plug into their
machines, especially for those running servers. That will guarentee
that someday Firewire will be replaced by something with a greater
capacity than Firewire will ever have.
So Firewire attacks SCSI which attacked IDE, and someday some new
connector is going to come out with pipes so huge that people who want
their I/O to scream are going to instantly switch over to this new
Firewire successor.
Granted, the swapover will be painful, but people needing those
devices and that speed are likely to make the investment in a new
connector technology for the gains that it will bring. To swap off of
USB would require similar gains and given USB's niche, that ain't
likely to happen.
Odds are we'll see a USB-2, possible with some superior data
transmission protocols, but it won't be accepted unless its backwards
compatible (or there's a simple adaptor plug available). But the
evolution of USB given its modest target is going to be very slow.
In short, Firewire is going to attack SCSI which is attacking IDE, but
USB is going after serial, parallel, and the like. The two are going
after whole different breeds of periphial and so are not in direct
competition with each other.
Technical merits aside, the currently installed base alone is enough to keep SCSI going for more than two years. Don't hold your breath waiting for Firewire RAID.
SCSI is unsurpassed in dealing with multiple peripherals on one bus because of the intelligence of the SCSI protocol and controller cards. If you fill a FireWire line to the max with devices, the CPU will run out of cycles to service them all. By contrast, if you fill a SCSI bus to the max, all your devices hum along smoothly because the SCSI card manages everything well.
I agree that FireWire has some incredible advantages which will help it dominate the consumer market. However, SCSI will survive in the high end, not just (as you say) the very high end. Many servers require the intelligent I/O handling that SCSI provides. It's not just the NSA that needs SCSI.
Off topic points:
Many people have put forth passionate arguments explaining just how much better FireWire is than SCSI. These same arguments are just as damning against IDE, if not more so. Combined with the fact that FireWire is targeted at the "low-end" consumer market (which is the only market that uses IDE), it seems absolutely incredible to me that consumer markets would reject SCSI on grounds of being inferior to FireWire, while still accepting IDE.
I admit that IDE has a rather large installed base. However, at least one person here has argued that the size of the SCSI installed base is no defense against FireWire. Turning that argument around, I posit (for the same reasons) that the size of the IDE installed base will not prevent IDE from being supplanted by FireWire.
I hope that my remarks have convinced some people that at least one of Swanson's statements is wrong.
As some have put forward, SCSI's still got an edge on things on the high-end. Only things that could put a real crimp in SCSI are Fibre Channel and IEEE-1355.
For those who don't know what IEEE-1355 is, it's a high-speed switched networking system that is derived from the Transputer's parallel processing link system. Baseline spec is capable of ~300Mbps with the high end being in the 1-3Gbps range. The parts cost for the basline adapter is about the cost of an RS-232 port, and the switches are on a par with the USB hub in complexity and cost.
The snag? These devices are still in their infancy- there's not a lot of the low end stuff out there and the high-end units are engineering prototypes that were built to make high performance clustering possible.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Firewire sounds so fast because of the way they quote the transfer speeds. Since Firewire is a serial protocol, they use Mega Bits per second. And since SCSI is a parallel bus, SCSI speeds are quoted in Mega Bytes per second.
Currently available SCSI:
-----------------------------------------------
SCSI-1 : 5 MBytes/sec : 40 Mbits/sec
SCSI-2 : 10 MBytes/sec : 80 Mbits/sec
SCSI-2 Wide : 20 MBytes/sec : 160 Mbits/sec
Ultra Wide : 40 MBytes/sec : 320 Mbits/sec
Ultra2 Wide : 80 MBytes/sec : 640 Mbits/sec
SCSI next year:
-----------------------------------------------
Ultra3 Wide : 160 MBytes/sec : 1280 Mbits/sec
Currently available FireWire:
-----------------------------------------------
Standard : 12.5 MBytes/sec: 100 Mbits/sec
High Perf. : 25 MBytes/sec : 200 Mbits/sec
FireWire next year:
-----------------------------------------------
Higher Perf. : 50 MBytes/sec : 400 Mbits/sec
FireWire some day:
-----------------------------------------------
Ultra Perf. : 150 MBytes/sec : 1200 Mbits/sec
So there you can see that the numbers being thrown around regarding FireWires blazing speed can be a little misleading. Today's SCSI drives are much faster than today's FireWire, and will continue to be so in the future. Any time a high performance server is being built, SCSI will remain the storage interface of choice for the forseeable future.
Now if all you want to do is hook up that nifty new camcorder or color scanner to your PC or Mac, FireWire is definitely the way to go.
Regards.
"The only good windmill is a tilted windmill."
And of course, a single drive can't pump data through a slow FireWire bus any faster than it can a fast SCSI bus. ;^)
"The only good windmill is a tilted windmill."
Regards.
"The only good windmill is a tilted windmill."
Change SCSI speed references to MBits/sec instead of MBytes/sec. Then we wouldn't hear so much about FireWire's "blazing" speeds.
Who would want crummy old 400Mbps FireWire when you can get 640Mbps Ultra2 SCSI? It's marketing, man. Perception is everything.
Regards.
"The only good windmill is a tilted windmill."
How good could Firewire be if it allows hooking a crappy IDE drive to it?
If it works--great! SCSI drives are too expensive. My 2.1Gig UW still costs about the same now as when I bought it two years ago. It's better if everyone buys what I buy. No need to be that rebelious.
But what I don't understand is why SCSI is being bypassed (yet again). It's going to be a while before Firewire get up to SCSI speeds. Why not just concentrate on SCSI?
Weird, but it's the best reasoning I've heard.
Chris
Except that SCSI is still faster (by far) than Firewire. If anything, it seems that Firewire will attack IDE, and Fibre Channel will attack SCSI.
--
Aaron Gaudio
"The fool finds ignorance all around him.
"Every man is a mob, a chain gang of idiots." - Jonathan Nolan, Memento Mori
A href=
"http://www.lucent.com/micro/NEWS/111098a.html"
http://www.lucent.com/micro/NEWS/111098a.html
I seriously doubt that firewire will replace scsi in servers within 2 years. Server designs usually take a bit longer than desktops to use "trendy" features. Perhaps support for firewire will be introduced in addition to scsi over the next few years. I know firewire has been around for sometime now, and I'd sure like to try it out. But I don't see scsi disappearing.
Well, one acronym, actually: HIPPI.
Anyone have HIPPI devices on their PC? How about at work? Not very common, huh? Assuming SCSI survives and continues to evolve, it'll be the HIPPI of the next century, IMO. Here's what I base that on:
Like any transition, it'll take time. Just look at how long it took for ISA to disappear! SCSI will march on in some form, but FireWire is clearly the future of the vast majority of the market (i.e. home and "normal" business users). It's my hope that the coming generations of the FireWire spec will surpass SCSI so even the high-end will get the benefits of serial interface simplicity. Now, about that analog video connector... ;-)
SCSI comes at a high price mainly because most people buy IDE shit - and of course that results in a vicious circle.
DMA IDE interfaces have been around for a while - that's why there's the "Ultra" in UltraDMA to indicate that it's a faster DMA mode.
It's not as simple as that. IDE was introduced as a backwards-compatible replacement for ST506-type controllers, offering better performance at around the same price. The original IDE interfaces were simply buffered interfaces to a subset of the ISA bus. IDE drives emulate the registers of an ST506-type controller, but with an extended command set. This means that no changes to BIOS or OS code were needed to support IDE drives initially.
All the backwards-compatibility hacks that are involved in today's drives makes them pretty complex and should make them more expensive than SCSI drives. However, that compatibility allowed IDE to sweep the PC market in the early 1990s and meant they could be produced in huge quantities, and hence more cheaply than SCSI drives.
Since IDE took over the low end of the market, most SCSI drives are high-end products with additional features over IDE. This increases the price differential further.
Alot of you asutely point out that SCSI can be faster than firewire. But can you hot plug SCSI? Is the "wire" thin and flexible? Do you need to worry about termination, and IDs with firewire? Is the SCSI connector designed for 8 year old boys ( game boy? ). Firewire will an awesome way to connect things besides diskdrives. Digital cameras? ALL DV camcorders have 'em. Scanners? Please broaden your though processes just a bit.
Well, some of us can, but you sure can't.
40*8=240megabits
Really? I get 320.
Apple's not the first to make firewire a standard interface. Sony's VAIO computers (portables and towers) have firewire on board too.
Does anyone know anything about it though? For example, is the iLink on the Sony portables going to support Castlewood's orb? Or is it some kind of special Sony version?
\
scsi will die and ide will live on?
are we living in an insane world?
On a long enough timeline, the survival rate for everyone drops to zero.
If Firewire gets as cheap as they say it will, they will compliment each other.
Firewire on the low end, Parallel SCSI in the mid range, and Fibre Channel on the high end.
The best part is all three interfaces use the same protocol. The same SCSI midlayer and higher level drivers can be used for all three. You just need to write the low level driver for your particular interface card.
Fibre Channel isn't parallel. It's serial in both its copper and fiber implemetations.
It's just a lot faster than firewire.
As I under stand it, Fibre channel is the parallel version of Firewire. Also, Fibre Channel can be fiber optics, which boosts speeds as well.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
SCSI is way overpriced and it is just waiting for something to topple it. Sure it will be around for a while just like ISA Bus, but it is on its way out, just like the ISA Bus. BTW UltraDMA is doing just that, easing the burden on the CPU, and this trend will continue in the future: IDE will get faster and more independent. This happens for really one specific reason - it is CHEAP, CHEAP, CHEAP.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
SCSI sucks because people building servers expect to pay big bucks for storage. THAT is why SCSI prices are still WAY above IDE, not because spindle RPMS are higher, or manufacturing costs higher or anything else. The manufacturers KNOW they can charge high prices for there $hit because people will pay for it. Of course prices have gone down since the beginning of SCSI, but they are not really scale for the performance increases compared to other computing technologies.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
I was directly responding to the previous poster by saying that
1. IDE will not be gone that soon because it is just so darn cheap.
2. SCSI is so expensive only because people are willing to pay for it.
3. SCSI is just waiting for competition
4. SCSI will then fade out like ISA
5. I have a free clue for him (and you) because he had one for everybody else
Maybe you should pay a little more attention befor wasting your time with diatribes that really miss what the discussion was about.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
Free clue to myself.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
I didn't say SCSI didn't have it's merits. On the contrary, it certainly does. I am talking about pure pricing scales.
I don't want to sell anything, buy anything, or process anything. I don't want to sell anything bought or processed...
AS far as i see it SCSI is not going to go away in 2 years. Ya sure its a little more expencive (actualy a lot), but people going for shear speed and no overhead will shurely stick with SCSI. in fact im going to buy a new SCSI drive today.
however for external use that would be great, not everone has a SCSI controler for you to bring your HD over and copy some data. then again not everone has firewire. If they implement it on all the MainBoards going out in the near future then i can see it becoming a trend. I have a laptop with a 10/100 NIC in it for my portibal needs. This sounds like USB to me, its hasnt taken over my serial devices yet, nothing beats my 8$ mouse, and is thier even a USB ISDN device made?
P.S. yaya i know my spelling sux
-magister-
If you think IEEE-1394 connections is going to replace SCSI, I kind of doubt it.
Ever heard of this thing called Fibre Channel?? It's a super-high-speed interface that makes Ultra-Wide SCSI seem slow in comparison. With the price of Fibre Channel devices starting to fall, I expect within a few years the average server will have Fibre Channel connections for the hard drives at least.
Raymond in Mountain View, CA
I think the reference to Firewire supplanting SCSI is in the area of new installations. With higher throughput and supposedly less cost there is an opportunity.
USB got screwed over because everyone waited on Mickeysoft to release software to use it. Basically think of it as being announced for primetime too early.
Apparently Jobs has made Apple brave again, going to new technologies quicker than the stumbling Wintel monopoly. If they can get some manufacturers behind Firewire (it seems they have a few) maybe this will help bring SCSI down to a reasonable level, instead of the price-fixed level its at now. (there cannot be much real price difference between IDE and SCSI drives as in one brand I know for sure the only difference is the circut board attached to the physical drive - yet the price of the drive is 200 bucks more)
..
. * Did aliens forget to remove your anal probe?
Methinks that comparing Firewire to Fibre Channel is a bit like comparing Windows and Unix, or modems and ATM. Both have their area of speciality.
:)
I work with high-end enterprise digital asset management systems - unix servers (generally Sun X500s) with RAID online storage(often A5000s) and HP MO jukeboxes providing near-line storage (up to the 1200ex, which stores 1.2 terabytes).
Sun's A5000s are FCAL devices and until something better than FCAL comes along, I'm not switching. From what I've seen, Firewire isn't better; it's just had good PR.
Anyway, everyone seems to have forgotten that both Fibre Channel and IEEE P1394 are part of the SCSI-3 standard.
Dodger
We have a pile of Fibre Channel connected disk right here where I work. The product is the Sun A5000 arrays. At a relatively low level the fibre channel is some special high speed bus. But when you install the drivers in Solaris the fibre channel connected disk is presented to the operating system as a pile of SCSI controllers, targets and disks. It seems to me that regardless of the fact that the underlying bus is fibre channel, fire wire or smoke signals atleast one layer of abstraction is going to be SCSI.
The first SCSI drive whose purchase I was involved in cost something like $400 for 20MB. A couple of years later I was buying 200MB for $350. Just a couple of weeks ago I was able to get 4.2GB UW disks for under $250. (Of course, I still pay nearly a grand for a 4GB StorageWorks disk; I guess that little plastic module costs quite a bit to manufacture)
How is SCSI not getting cheaper? There are advantages to SCSI that the people who buy IDE apparently don't need to take of advantage of (performance in multitasking environments, etc.). If you're going to need lots of disk space, IMHO, SCSI is the only way to go unless you want to keep replacing your four IDE disks with larger and larger sized drives (and somehow dealing with the risk of having more and more data on a single point of failure^H^H^H^H^H^H^H^H^H^H^H^H^H^H^H^Hspindle).
I do wish controllers would come down in price. I'd much rather use a few fewer drives on a controller and use multiple controllers to balance I/O and having multiple controllers is The Only Way To Go (tm) for drive mirroring.
CUR ALLOC 20195.....5804M
Will SCSI ever achieve 80MByte/sec? You seem to be quoting speeds from something that doesn't exist yet.
More importantly, SCSI still has that 7-device limit. That is where FireWire really scores over SCSI.
--
Don't imitate. Enervate.
Agree that Device Bay is probably key to Firewire's success. (Those who don't know, Device Bay is a spec which would allow drives to be universally snapped in, hot, with no cables, screws, jumpers, etc.)
I thought 1394 has a standard plug so that we won't have go go through the connector hell that is SCSI. (At least 5 external plugs by my count.)
Business. Numbers. Money. People. Computer World.
Keep saying it, but you might be dead wrong.
Sony has plans to make Firewire a univeral interconnect for stereo systems. Just replace that ratsnest of RCA plugs and digital interconnects with one firewire chain. Sounds like a good idea to me.
(Agreed that a Firewire keyboard will probably not see the light of day, tho.)
Business. Numbers. Money. People. Computer World.
Since any nine year old can connect gameboys together, and presumably not bend the pins or gunk up the cables, I don't see how this is a bad thing!
Especially when compared to the expensive hell of different SCSI cable specs, connectors, and terminators.
Business. Numbers. Money. People. Computer World.
hahahahahahahahahahahahahahahahahahahahahahahaha!
Yeah Apple is full of bright ideas lately... firewire is sooo great! that's why they didnt put it on the imac. death of SCSI? I dont think so..
EIDE versus SCSI.... lemme see... IDE sucks, SCSI rules. IDE is slow, SCSI is lightning fast. IDE is a limited bus that is prone to problems, SCSI is unlimited and is self correcting. IDE forces the computer to do all the work, SCSI takes the load off the processor. I see firewire and usb dying.. SCSI will live forever unless intel morons try to kill it.
Another reason why we all dont have mac's instead of Intel based computers.. Apple cant make a good decision to save their lives.
Do not look at laser with remaining good eye.
people making servers are charging extra for scsi.. simple solution... MAKE YOUR OWN SERVER!
a server is no more complex than a desktop, any IS person worth the paper bag they were shipped in can do it. (Unless they are MSCE then they dont know what SCSI is or what a bios chip looks like) DELL servers are crap, for the same price an IS department can build out of premium components twice the machine, instead of the second hand crap that dell sells.
Do not look at laser with remaining good eye.
What an intilligent decision! Squeeze data through a serial protocol! good for toys like tape backup or joysticks.. as for HDD's.. let's get a 64bit wide scsi bus. and add more processors to the hardware so we dont waste cpu cycles doing the work that the prephrials are supposed to do! the scsi card had to do all the data work it's self, the HDD's have to deal with their own crap on theri own and leave my CPU cycles alone! If hardware manufacturers would get their heads out of their butts we wouldnt need 9950Mhz computers with 100 processors to run our software because the dang hardware requires complex drivers. USB is a joke. Firewire had potential. EIDE is for the poor and wretched. Gimmie SCSI Ultra Wide! and prephrials that are computers on their own!
BTW, I proved this point to my friends last week, a 586 p100 running quake II quite nicely.. All prephrials, video, hdd,sound,etc.. were of the self processing type.
BAN any hardware that requires software to run them!!! (other than simple drivers)
Do not look at laser with remaining good eye.
If you put enough devices in parallel, you can get darn close to 80MB/sec. I setup a system recently with 6 10,000RPM drives in parallel that had a sustained transfer rate of very close to 80MB/sec. Granted, the individual devices won't do much over 20MB/sec but when you run them in parallel with a good controller you can do amazing things.
At the time, I was doing a test of running SCSI Ultra2 head to head against Fibre Channel. It ended up being a very close race which made it quite hard to justify the expense of Fibre Channel.
World Beach List, my latest project.
Actually, there already is a project to get USB running on Linux. They already have support for some devices. Sorry, but I wasn't able to dig up the URL.
World Beach List, my latest project.
SCSI will be with us for a long time. Fine by me... it's mature, stable, and plenty quick for 9/10 of the things that need to be done with storage. I'm all for Firewire and crew, but I just don't see the need to rip out my SCSI installed base yet. It does what I need it to do.
-- ultra1
You all are missing a large part of the reason firewire is so great. If you would have watched the demo at macworld on ZDTV (someone posted the link on here somewhere) you would see that you can chain computers together with devices.
For instance, lets say I have a scanner, I can hook that scanner to computer1, then hook computer1 to computer2, computer2 to computer3, and on down the line, and ALL the computers can share that scanner as if it was locally attached to each one.
The same can be done with a hard drive, the computers will think it is local.
I can have an external hard drive at the office, bring it home and hotplug it into my home computer and it will spin up WITHOUT a power adapter and show up on my screen in a few seconds.
I think that firewire can give up to 60watts of power, so you don't have to worry about plugging most of your devices in.
"A great deal of intelligence can be invested in ignorance when the need for illusion is deep." --Saul Belloe
Well I don't know much about Fibre Channel so, correct me if I'm wrong, it is much more expensive than FW, SCSI, and Ethernet. Maybe this will be a viable alternative in the future when it is cheaper, but right now it isn't.
As for ethernet, I'm not aware of any scanners, hard drives, digital cameras, or any common peripherals beside printers that have ethernet ports built into them.
"A great deal of intelligence can be invested in ignorance when the need for illusion is deep." --Saul Belloe
I expect that the drive manufacturers don't give a damn whether they sell firewire or SCSI drives.
On the other hand, I think you're right about controllers. If an NCR 53C875 controller goes for Cdn$115 these days, there's no reason Adaptec should be charging Cdn$250 for their card.
cjs
The world's most portable OS: http://www.netbsd.org.
SCSI use to be horribly expensive. It's a lot more reasonable now (I can remember not too long ago SCSI cards costing $500). There used to also be a significant premium for SCSI disks; that's gone, as you can find general parity in pricing for EIDE and SCSI disks (although some places insist on charing more). Used SCSI equipment is often cheaper than comparable IDE equipment.
If FireWire be a better technology, then I do hope it will take over SCSI. However, it needs to be free of choking patents and the specifications need to be very open if we ever want to use it with an opensource operating system.
--jon. Postel is dead. May we all mourn his, and our, loss.
You're telling me. Adaptec UWs have been stable in price for a long time (hovering at around $160 +/- $10 for OEM). If firewire can kill scsi device prices, I think we'll see more people turning to scsi as not only a viable storage option, but also an inexpensive storage option.
one application which needs arrays of big
drives with big bandwidth is uncompressed
video editing/effects (which, coincidentally,
happens to be the bread and butter of the
company i work for). for example, real-time
uncompressed video requires about 30MB/sec
(or 240Mb/sec).
however, for a simple real-time dissolve
(2 read streams and one write), triple it
to 90MB/sec, and you're beyond what a
single UWSCSI channel will do.
real-time film-res playback
(2Kx1.5Kx48bits colour) is about 340MB/sec.
i'm sure there are other applications, such
as serving many streams of high-quality audio
or video over high-bandwidth net connections
which require striped drive arrays connected
over multiple SCSI channels as well.
I think people need to realize that there is MUCH more to the computer industry than than the consumer market. YES, FireWire is a nice alternative to SCSI for a large number of devices. And yes, I'll be one of the first people to note that SCSI has many many issues, and will be phased out eventually.
But, let me be the one to tell you right now that the industry has absolutely NO interest in making hard-drives for the FireWire architecture. It is GREAT for digital cameras and high bandwidth applications, but is NOT tested in hard-disk storage to the extent of SCSI and FibreChannel.
FibreChannel was conceived to be the replacement to SCSI, and the industry has placed too much time and money into its conception to abandon it so readily, as some suggest. So, as it would be nice to have one standard interface for all computer devices, I seriously doubt it will be FireWire, but, rather something yet to be conceived.
Thanks for your time
John Ehn
Ehn Consulting
"FireWire" is a buzz-word amongst the posters on this forum but they fail to see that it is nothing but a physical transport for protocols just like Ethernet is for network protocols.
The currently approved SCSI standard (SCSI-3) is currently capabable of using four different physical transports including the old parallell interface and FireWire IEEE-1394.
Check out the T10 homepage for further information.
http://www.symbios.com/x3t10/scsi-3.htm
--
Mattias Sandgren
Umeå, Sweden
>>Your average drive is probably no faster than 15MB/sec at best. This is true of both IDE and SCSI drive
Dude, first of all, I don't know many drives that are _really_ capable of sustaining 15MB/sec long enough to really call it a 15MB/sec drive. A Micropolis Stinger Ultra Wide SCSI @ 5400 RPM's gets around 7.8MB/sec on large files(adaptec SCSI bench). 33MB/sec on UDMA is the burst rate. How often is this achieved? Not often. My old Seagate 1220a EIDE drive (rated at 13.3 MB/sec burst) rarely got more than 900KB/sec under normal conditions.
>>80MB/sec. Who needs that much bandwidth when the drives are only capable of 15MB/sec?
Because with SCSI, the bus can handle _simultaneous_ I/O to and from the devices. This adds up when you have 2 hard drives, a cd-RW. cdrom and zip all on the same bus.