Ok, so I have an opportunity to aquire several SCSI drives.
I know bupkis about PC innards, to be honest, but I seem to recall reading somehwere that SCSI drives have to be paired, as in 2 80g drives, or somesuch.
Is that the case, or am I thinking of something else entirely?
Thanks for the assist!
What kind of SCSI drives? There’ve been many standards.
The pairing is only for RAID. You can RAID non-SCSI drives too. You will need a special RAID controller. Pairing does make things twice as fast though.
Usually, there is no benefit to SCSI drives, although they do tend to be made in higher RPMs than SATA/IDE drives. (Eg, 15k RPM.)
Chances are, you shouldn’t buy them. And if you try to use them, you’ll need to buy additional hardware. There’s many options in getting very fast storage. E.g, 10k Raptor SATA drives or even SSDs. Then again, I don’t want to spoil your excitement…
The SCSI drives worth having you can’t afford. The SCSI drives you’re aquiring are probably old and slower than the $100 special you pick up at newegg.
That said: You need a SCSI card, and SCSI cables, and you need to deal with termination, and possibly setting the ID on each drive.
In short, it’ll be slower AND cost more than just going with consumer stuff that’s already most likely on your motherboard.
Then there’s type. They was IIRC SCSI, SCSI-2, Wide SCSI, Ultra SCSI, Ultra Wide SCSI…and currently, Servers use SAS or serial attached SCSI. There’s an iSCSI in there too.
If you were acquiring an older server, WITH the drives and adapter and such, I’d say go for it as a science experiment. If you have to recearch even one aspect, I’d pass.
FWIW: SCSI - Wikipedia ETA: my SCSI types listed above a woefully incorrect, refer to the Wiki.
No there is no pairing. Its just a dumb drive. Your first problem will be that you most likely do not have a SCSI interface on your computer. Youll need to buy a SCSI card. Thats an easy 50 or 60 bucks there. Whats the capacity and age of these drives? You can get a super-fast 500gig SATA drive for like 70-80 dollars right now, new. The performance out of my WD 500gig drives is pretty impressive. Much higher than a SCSI drive from a couple years ago. I doubt inheriting these drives really is worth the effort.
Maybe the linear transfer rate is fast, but that’s not the most important factor in performance. A 15k scsi drive will have much better seek time than your 7200 rpm drive. Much, much better. Of course there are still negatives. It will also be much, much louder.
SCSI drives evolved out of the DSSI drives in Vax computers. ATA (also called IDE) drives developed from the original ST 506 disk drive interface from IBM AT computers (ATA means “AT attachment”, and IDE means "Integrated Drive Electronics, i.e. the ST 506 style disk controller is built into the drive itself).
DSSI drives were built for Vaxes, which were usually used as servers. You could have a whole bunch of disk drives all attached to one server. ST 506 style drives were designed for a PC and could only have two drives per interface. These uses and limitations evolved forward with the drive interfaces. SCSI drives were designed for multitasking computers, and you could put a whole bunch of the together on one long line. IDE drives were for made for single user computers, and you could only have two of them per interface.
Fast forward a bit, and SCSI drives become really popular for servers. They cost a bit more, but the way that the drive interface is designed, they just work faster in servers than ATA drives do. In a single user environment like a home PC, ATA is just as fast as SCSI most of the time, so there’s no benefit to home users for using SCSI. So, SCSI becomes standard for servers, and ATA drives become standard for home PCs.
SCSI drives don’t have to be paired, but since they often ran in servers, and servers need to run 24/7, they created these things called a “RAID” (Redundant Array of Inexpensive Disks). Basically, you take two identical disks with a special controller, and run them together as one disk. Each disk contains an exact copy of what the other has, so if you have two 80 GB disks in a RAID, it looks like one 80 GB disk, not like a 160 GB disk. The idea here is to have two copies, not to make a bigger disk. If one of the disks break, you just use the other one alone until the broken disk is replaced. Your disk went POOF but your server stayed running. RAIDs also tend to be slower than single disks, because when you write the data out to the disk, you have to write it to both disks. Reads are just as fast, though, since you only need to read from one disk (each disk has an identical copy, so it doesn’t matter which one you read from).
RAIDs cost money. Basically it’s twice the cost in disk drives for the same amount of storage space. Servers need them to keep their servers running though, so you see RAIDs in servers a lot but you don’t see them in home PCs a lot. Since servers used to use SCSI a lot more than ATA, you would therefore see SCSI RAIDs a lot more than ATA RAIDs. ATA RAIDs did (and do still) exist though.
This is probably why you got the idea that SCSI drives had to be paired. It was common in servers, but isn’t necessary. You can just as easily run SCSI drives as single drives.
These days, there have been tremendous improvements in both the disk interface and the disks themselves. With data caching and faster interfaces, the performance difference between SATA and SCSI isn’t as big as it once was. SCSI still has a large presence in the server market though. The really high performance drives in top of the line servers will often have a SCSI interface. This is why Unintentionally Blank said the SCSI drives worth having you can’t afford. The ones that really have any performance benefit are the really expensive ones designed for great big expensive servers.
A PC typically doesn’t have a SCSI interface in it these days. SCSI controllers can also be somewhat twitchy to get operating correctly. If your SCSI drives come with a RAID controller, these can be particularly painful to set up sometimes. If you are getting all of this stuff for free or pretty cheap then it might be worth it. If you are just getting the drives with no controller, it’s probably not worth messing with. You can buy an SATA drive cheaper than the SCSI controller, and you’ll likely spend an awful lot of time setting up the SCSI interface to get it to work properly.
Also, like ATA, there’s different types of SCSI. If you get the wrong controller card it will be like trying to plug an SATA drive into an ATA parallel interface. Ain’t gonna work. Worse, some versions of SCSI have different connectors for the same interface, so not only do you have to match up interfaces but you also have to match up connectors. Then of course you also need terminators, which ATA drives don’t have to mess with. SCSI drives are nice with respect to the fact that you can chain a whole bunch of them together, but you need a terminator at the end of the line, even if you only have one drive. Some drives have built in terminators which you can enable with jumpers, others don’t.
Even assuming these are 15k drives, I really doubt typical personal usage (web, email, gaming) will prove to be noticeable a upgrade. A single user system is better off with higher linear speeds, especially reads. Not to mention used drives are always bad luck. You dont know how hard theyve been worked or how much life they have left. Unless hes building a file server thats going to be pretty busy and has a box with a SCSI card, then hes better off with a new SATA drive.
That was always a bs artifact of benchmark programs that were specifically designed to not take the huge, nondeterministic variable of hard drive performance into account.
Fact is, most times your computer is grinding to a halt, the hard drive light is blinking like mad. And when it’s doing that, it’s the random access performance that matters most.
Actually, the seek time is not improved, but the latency is.
How might those differ?
Actually, if you REALLY want a bump in performance, look into an SSD drive. The lack of any moving parts GREATLY improves random access.
This is a two year old video, current drives have been improved:
Seek time is the time for the head to slew to the correct track; latency is the average time for the data to arrive under the head once it is in position.
That might be an interesting historical definition, but today hard drives only quote “seek time,” and it goes down significantly with speed. Eg, typical seek time for a 7200 RPM disk is 7ms, while for a 10k Raptor it is 4.2ms. But perhaps you are onto something, that seek time has more to do with the motor being used for the arm than the motor being used for the platters.
Distressingly, btw, manufacturers seem to have stopped quoting seek time (or latency) for their consumer drives. Western Digital and Hitachi both don’t have it in the specs for most drives. Western Digital doesn’t even specify platter speed for their lowest-end ones. Wtf.
I remember when the SCSI interface for PCs was “new”, and everyone just knew they were going to take over the personal computer world. You could put eight SCSI drives on a single controller! And most of the access logic was on and processed by the controller rather than handled by your PC software! And technically things other than hard disk drives could be added to the SCSI interface! And sure, SCSI cost a bit more than IDE but everybody was certain that the hard-core PC user would be willing to pay that bit more for all of those extra advantages!
(Sigh) and I bought into the hype.
One of the reasons they stopped quoting seek times and latencies is that the numbers have gotten really hokey. In the old days, you requested data, the drive moved the head, read the track, and gave you the data. Modern disk drives have a lot of cache memory inside of them. Not only do they buffer up data, but they have prediction algorithms and try to buffer up data that you haven’t even requested yet. The result is that when their prediction is right, you get these ridiculously low latencies that make the drive seem unbelievably fast. Once you exceed the drive’s buffering you end up going back to the old limits of track to track head seek time and how quickly you can read a sector (faster rpm equals faster sector reads), which is significantly slower.
They haven’t yet come up with any standardized ways of measuring disk drive latencies for these newer drives that all of the different drive manufacturers can agree on.
Disk drives are arranged into tracks and sectors, which, if you could see them, would look like rings on a tree. The way a disk drive reads and writes is that it first moves the head to the correct “ring” (track). Then as the platter rotates under the head, it reads the sector off of the track, or in the case of a write, writes the sector out to the track.
The “seek time” is how quickly the drive can move the head to the right track. This depends on the design of the head, the coils that move it, etc.
How quickly you can read the sector depends on how fast the platter is spinning under the head. The faster the RPM of the drive, the faster the track read or write occurs.
The overall I/O latency of the drive is the seek time plus the sector access time.
Making the drive spin faster reduces the total I/O latency, but doesn’t do anything for the seek time.
Likewise, improving the design of the head without changing the speed of the drive will improve the seek time but won’t affect the sector read time.