FASTEST SETUP PERIOD 15,000rpm scsi or 10,000 SATA?

lb19984

Distinguished
Nov 18, 2003
671
0
18,980
Hi everyone havent posted here in a while.... been almost a year since ive had a home-buit computer and it looks like i have some extra cash to built a new rig.

I currently have an old 74GB raptor (8mb buffer)
I have acess to the newest or some of the newest Server hard drives....(scsi 15,000 rpm etc
but from what i have reseearched in the past the would ultimately be slower because of a raid controoler? etctra card? something like that?
really not sure

So the question is what to get.. i will have 1 hard drive with OS on it and all my files on another HD
So the question is what will be the fastest?
newer 74gb rapor (16mb buffer)
150gb raptor
or a scsi Hard drive?
ive been out of it for about a year so sorry to be so newbi-esk
thanks in advaced!!!!!!

I will be running this with the top of the line asus mobo
good corsair memmory and the new 4 core cpus (whenever they come out) all overclocked like crazy :D :D
thanks again
 

lb19984

Distinguished
Nov 18, 2003
671
0
18,980
eeeeeeeeeeeeek i think this has been asked before
but in general when i put a raid controller card or whatever i need to be able to use scsi Hd's will that ultimately make the HD work slower than a SATA II raptor?
 

PCcashCow

Distinguished
Jun 19, 2002
1,091
0
19,280
Apples and oranges.
The Raptor trumps SCSI in two ways.
1. It's cheaper
2. Ment for Single user setups

But as far as looking into what will be ultimatly faster, SCSI would have the upper hand in most test, but marginal in some.
What you need to focus on is having a good controller card that has plenty of write back cahe on it and a speedy interface. PCIe (x8)SCSI cards are expensive and hard to find becuase the 68pin insterface is slowly ageing off. The LSI 320-2e is a great card, but when looking for one stay way for ebay. Youll only find Dell Perc4e/DC cards that support W2k3 out of the box only and cannot be upgraded to a true LSI Card. If your going to look into a PIC-X card for your system be sure that you have a PCI-X 133 bus. If not and your using PCI 2.0 your performance goes right out the door. SCSI can and does beat the newer Raptors but it's more exspensive and harder configure.
 

mkaibear

Distinguished
Sep 5, 2006
678
0
18,990
SCSI disks aren't optimised for single-user performance. They are tuned for maximum performance in a server usage pattern (lots of localised reads and writes, mostly contiguous data).

SATA drives (including the Raptor, despite its server origins) are tuned for single-user performance.

If you want the *absolute* top sequential transfer rate, go for a pair (or more) of 15k SAS or SCSI drives (I'd recommend SAS) in RAID 0.

If you want a faster desktop performance (which is much more important to most people, latency trumps STR in almost *all* aspects), go with Raptors. If you can afford it, RAID 0-ing them won't seriously hurt you.
 

lb19984

Distinguished
Nov 18, 2003
671
0
18,980
no i don not have the hard drives yet but i will mainly need them for regular PC usage... i am not completely sure what is best guys. maybe i can check what scsi hard drives i have acess to and then maybe i will be able to figure out if the raptors or the scso drives will be fastest
thanks alot
 

mkaibear

Distinguished
Sep 5, 2006
678
0
18,990
>If money is no object, get SCSI

Not if it's for general desktop use. Yes, you'll get the fastest transfer rates, but they're really not designed for desktop use.
 
>If money is no object, get SCSI

Not if it's for general desktop use. Yes, you'll get the fastest transfer rates, but they're really not designed for desktop use.

Most SCSI drives would work fine in desktop applications. Designed for server use normally just means better made higher quality kit. Why would you see this being a problem?

SATA would probably be the easier option in this instance as it avoids having to buy a controller card. Given the high price of raptors I wonder if this would be negated if the OP has access to SCSI drives for low/no cost.

How about a pair of the new 320Gb WD drives in RAID0? This would surely challenge (might not beat) a single raptor. Also give a much lower cost per gig.
 

niz

Distinguished
Feb 5, 2003
903
0
18,980
Apples and oranges.
The Raptor trumps SCSI in two ways.
1. It's cheaper
2. Ment for Single user setups

Agree with 1. but 2. is baloney.
My last company just spent 15 million on a new data center server with a giant SAN array and guess what drives it has (as also reccomended by the HP professional data center team). That thing is seriously multi-user.
 

mkaibear

Distinguished
Sep 5, 2006
678
0
18,990
>seriously multi-user

Actually, it's tuned for desktop use - WD say so.

What applications is your SAN going to supply data for? That may have had a bearing on them recommending Raptors.
 

mkaibear

Distinguished
Sep 5, 2006
678
0
18,990
>work fine in desktop apps

Yes, it will work fine - it just won't be as fast. It's tuned for contiguous data sets in a multi-user environment, not tuned for non-contiguous data in a single-user environment.

>challenge

It will have a better sequential transfer rate than a Raptor - it won't "feel" as fast, because it will have a much higher latency.
 
I dont think the difference would be that great. And dont forget plenty of SCSI drives earn there keep in the workstation area.

As for the RAID0 issues you are right about the latency issue although this is only one factor in the overall performance of a given drive. I know what you mean about 'feal' though.

I went and did a bit of digging on the performance of RAID0. This article seems quite interesting as it does some nice tests.
 

tirez

Distinguished
Dec 17, 2005
14
0
18,510
I would go Ultra 320 SCSI over SATA or SAS for the following reasons:

Interface Performance
-----------------------
Ultra 320 SCSI is capable of 640MB/s even with a single drive, you can measure this with a fast inteconnect bus(like pci-e), testing the burst speed. This is much faster than the 375MB/s of theoretical interface speed you can get from both SAS and SATA. The interface is able to handle up to seven drives per channel, for large setups.

Drive Performance
--------------------
Access time: An often overlooked performance metric by benchmarks like Sandra, but probably most important to performance in most situations.

Essentially, Access Time=Seek Time+Rotational Latency

A 15,000 rpm hard disk has an average rotational latency of 2ms, whereas a 10,000 rpm hard disk is 3ms. The seek time can be measured with a benchmark by measuring access time and subtracting rotational latency. Here is one place where scsi drives dominate, with seek times 50-100% less than competing SATA drives. Check Storage Review for benchmark results. Essentially access time is the time it takes the disk to spin to the proper sector + the time it takes the actuator to move to the proper cylinder to read or write data. Since operating systems and applications typically use hundreds of thousands of files spread accross the drive, this metric is especially critical with regard to overall performace accross the board.

Throughput: I would check out Storage Review's numbers on the throughput of the Seagate Cheetah 15k.5 drive to see the best numbers so far of any drive (both inner and outer cylinders). This drive doesn't post as fast of access times as the Maxtor Atlas 15kv2 though.

Keep in mind that arrays of drives add rotational latency due to the drives being out of synch with one another. Also controllers aren't necessarily designed for high throughput, but for data integrity. There is also the potential for controller overhead which can decrease throughput and increase access time.

Basically, when Storage Review did their reviews of the latest Raptor drives, they went through their typical benchmarks, and for some odd reason, decided to weigh much more heavily Veritests benchmark, over true dedicated disk benchmarks like IOMETER. In turn they decided that small file throughput is more important than large file throughput, and that the other benchmarks didn't represent true desktop performance. If you compare say, the Atlas 15kv2 versus the latest Raptor in their comparison tests, you will see that in all other tests, the Atlas15kv2 literally destroys the Raptor drives, sometime by margins greater than 100 percent(as file sizes get larger).

Update: If I were wanting to get the fastest storage today, my choice would be the Seagate Cheetah 15k.5 and a nice PCI-E SCSI controller, like my current LSI Logic 320-2E, without using RAID. I would then backup to some cheaper drive(s) like I do now.

TiReZ
 

belvdr

Distinguished
Mar 26, 2006
380
0
18,780
Where SAS shines is when multiple drives are used. The bus speed is shared in conventional parallel SCSI (Ultra320 and below). In SAS, just as in SATA, each drive has its own dedicated connection; nothing is shared, except the controller.

Now, when you take SAS vs SCSI in RAID arrays, SAS is able to keep up no problem. The beauty in the SAS drives is also their size and speed: 2.5" 10k. We did some benchmarking with Oracle and various other read/write tests, and found the SAS 2.5" 10k drives kept up with the Ultra320 15k drives. Of course, the controller was different, but that was the only difference.

Update: If I were wanting to get the fastest storage today, my choice would be the Seagate Cheetah 15k.5 and a nice PCI-E SCSI controller, like my current LSI Logic 320-2E, without using RAID. I would then backup to some cheaper drive(s) like I do now.

And why would single drives be faster than an array? Of course, some tests will be slower, but overall, I'd go with an array any day over a single drive (provided finances and cooling support it).
 

ericlecarde

Distinguished
Aug 22, 2006
9
0
18,510
If your lookin for pure speed you might consider gigabytes i-ram "solid state" HDD. I know its only capable of reaching 4 gigs but it pretty much destroys any SCSI or SATA RAID setup. Just a thought .
 

tirez

Distinguished
Dec 17, 2005
14
0
18,510
You said they kept up, but there would be no reason they wouldn't, due to the drives internal throughput being lower than the controller's capability. Try measuring burst speed and see what happens.

Go to: http://www.benchmarkhq.ru/english.html?/be_hdd.html and download HD_Speed to do this. Make sure you check both small and large file sizes, and test burst speed.

Burst speed isn't in an indicator of a drive's performance though. You can have a really slow drive on a really fast controller and you will only see the performance of the physical drive. It is, however, an indicator of the capability of the interface, and the peripheral bus.

Also, which particular drives did you test?

Addition: Raid increases the chance of failure of your storage susbsystem with each drive you add. I guess if you need really fast thoughput for some reason raid would be fine(like a/v work), but i would rather have a single 15k.5 with its max 123MB/S than any pair of Raptors, getting the huge seek time and rotational latency benefit.

Using the above mentioned program I am able to achieve 512MB/S burst speed on my LSI 320-2E (using max file size), this is due to my motherboard's second graphics slot only supporting 2x PCI-E lanes. I am hoping my next motherboard will have a full 8x slot with this card, so I can get the controller's actual 640MB/S speed, for the heck of it. I found the card pulled from a new system off Ebay for $150, but you can get Dell Perc 4e which is the same card, too.

TiReZ