SCSI vs SATA

[Amuthan Vethanayagam]

Hi Guys,

I want a clear opinion about SCSI and SATA when it comes to PC based NLE Systems.

Years ago I used to work with a 15000 RPM SCSI and I found it very comfortable.

As SATA is cheaper and the performance is better than parallel ATA i started using SATA.

At a ceration point I stopped using SCSI as the hunger for space became higher and the SCSI prices did not come down as expected.

Now I have a 10000 RPM SATA2 as my primary hard drive and 7200 RPM SATA2 Drives (2) on RAID0 (stripes) Array.

I'm not having the same comfort I used to have when I was using SCSI.
How many of you still use SCSI in your primary hard drive?

What is the impact of Virtual Memory when you use low capacity primary drives?

Kindly help me with some professional answers to clear my doubts.

Amuthan

[Harm Millaard]

The funny thing is I used to have SCSI drives, but the number of failures, disk crashes and inherent data loss turned me off. Also because of the price difference you mentioned, I turned to SATA's and have had a couple of failures, but luckily they did not result in data loss because of Raid5 configurations.

We all know that SATA disks lose performance when they get filled more than 50-60%. But the largest SCSI disk is 300 GB and a run-of-the-mill SATA has 500 GB. If you don't use that disk to store more than 300 GB you have the same capacity as the largest SCSI, but at a fraction of the price. Example a Seagate Cheetah 15K 300 GB SCSI is around € 850 and a Samsung Spinpoint T166 SATA 500 GB is around € 120.

Just yesterday I received a price quote for a dual Intel 5355 rackmount system with 8 GB RAM, redundant PS, 16 Hitachi 500 GB SATA's and an Areca 1260 Raid controller with 1 GB cache for around € 10K. Can you imagine the price tag with 16 Seagate Cheetah's in there?

In terms of performance SATA has caught up with SCSI IMO. In terms of reliability, I don't know. But if you configure your raids carefully, using Raid6 for instance, you can easily swap crashed drives without losing data and still save significantly on the $$ in comparison to SCSI.

[James Harring]

I am surprised Harm reports SCSI failures. I run servers for well over 4 years straight 24x7x365 with hardware based RAIDs and have relatively few failures. All drives can fail, but my experience is SCSI is most overall reliable.

SCSI is normally used on servers and storage area networks; high SUSTAINED bandwidth, better overall build quality and more robust overall as the manufacturers still consider them a "botique" item, not a commodity. This is why the price has not dropped that much; as a tech, if I am building a server, I don't care about price. I want reliability.

On the surface SATA/SCSI are neck and neck, but in sustained thruput SCSI is better.

However, I see no reason to use SCSI in an NLE these days. You can find excellent performance on a SATA RAID array. It's really about bandwidth which translates to the ability to deliver data from the disk to main memory. Both are above the bar for most all video work. I would qualify this statement by saying I've not evaluated it against uncompressed video, but anything less, should not be an issue to use SATA.

Be aware there are two main SATA config's 1.5 and 3.0 gbs currently. Be sure all your drives are the same "flavor." This reduces the chance a slower SATA is on the same bus as the faster one, thus hurting performance. Also be sure to verify they are running at the higher speed with tools on the manufacturer website. Sometimes it takes a registry change, software setting or driver update to make it "go fast." Be sure to enable Native Command Queuing if available is enabled too. Seagate's 7200.10 series is pretty good.

Another thing you may want to consider is a relatively new technology called e-SATA. It's not a drive, but rather a way to connect an external SATA drive to the PC. You create the array in an external box and plug into the pc thru a special connector. It's about 3 times faster than firewire.

>>What is the impact of Virtual Memory when you use low capacity primary drives?
Actually it is irrelevant -- as it's just a chunk of a hard drive set aside to act as RAM, so really doesn't matter. What does matter is size of your page file and where it is located. I suggest if not be on the boot volume, but on another non-removable drive*. (A partition of your main drive does not count, we're talking physical drive, not logical.). Rule-of-thumb is size should be 1.5 physical ram size. So 1gig ram, page file should be 1.5 gig.

I also noticed your are RAID0. This leaves you vulnerable to hard drive crash, everything on BOTH drives will be lost is one drive crashes. I suggest you consider adding at least one more drive to create a minimal RAID5. This drive is parity, so if one drive fails, it can be rebuilt. RAID0 is faster, but vulnerable. I don't see the trade off worth it. If you change to RAID5, from RAID0, backup everything on the RAID, as all data will be destroyed.

More on SATA here:
http://en.wikipedia.org/wiki/Serial_ATA#External_SATA

Final note: putting a hard drive on a USB interface is not recommended by me. It's ok for casual use, like copying files, but for dedicated video work, e-sata, firewire (400 or 800) is recommended for a removable drive.

USB is a shared bus and at the mercy of whatever other devices are on that bus. There is so much (IMO) fraud with the USB spec these days, I try to avoid it for performance related implementations. I call fraud because many manufacturers take the official specification and treat it as a suggestion. Particularly this extends to power supplies, as well as the actual bus components. SOme buses don't segregate channels and some power supplies cannot deliver the required power output. I've tested a number of these.

*Windows will always require a small swap file on the boot drive fo technical reasons. Does not count for this conversation

[Daniel Browning]

Originally Posted by James Harring

I am surprised Harm reports SCSI failures. I run servers for well over 4 years straight 24x7x365 with hardware based RAIDs and have relatively few failures. All drives can fail, but my experience is SCSI is most overall reliable.

Agreed. Perhaps he was cooling his 15K drives as if they were 7K? In any case, two recent reports found that "server" disks were not more reliable than desktop drives:

http://labs.google.com/papers/disk_failures.pdf

http://www.usenix.org/events/fast07/...tml/index.html

I like the tests they do at http://storagereview.com/

They found that the WD Raptor 10K SATA is pretty fast: 60-90 MB/s, so that's what I use. The new Cheetah 15K.5 has 20% higher throughput, but 300% higher cost.

[Marco Wagner]

Originally Posted by Harm Millaard

We all know that SATA disks lose performance when they get filled more than 50-60%.

Where are you getting this? As far as I knew the further you go from the center of the platter the faster the access...hence the reason for putting your swap drive on the last partition you create.


Example:
Transfer Rate:
Inner Zone: from 44.2 MB/s to 74.5 MB/s.
Outer Zone: from 74.0 MB/s to 111.4 MB/s.
Random access time: from 5 ms to 15 ms.

[Harm Millaard]

Marco,

Have for instance a look here: http://anandtech.com/storage/showdoc.aspx?i=2922&p=4

[George Ellis]

Originally Posted by Harm Millaard

Marco,

Have for instance a look here: http://anandtech.com/storage/showdoc.aspx?i=2922&p=4

This is not unique to SATA drives. SCSI will do the same thing as the drive fills. This is issue with seek and write.

http://hdtune.com/results/Seagate_ST336753LC.gif I did a quick look and you can see the same thing with this 36GB U320 Seagate.

[Marco Wagner]

Those are pictures of transfer rates. I see nothing on either of those links to indicate they filled the drives to near capacity and then tested them... If I missed something please show me, I'm not trying to be "mr right".

[Harm Millaard]

Marco,

Unless I have always misunderstood these graphs (and mind you, I'm an autodidact, so I had a fool as a teacher) the horizontal axis shows to what % the disk is filled with data. AFAIK SCSI does not suffer as much as SATA, although the example George offered does not directly support my statement. In that case the previous point I made about the economics of SATA versus SCSI will be even more valid.

[Marco Wagner]

I'm seeing MB/sec in blue over access time in yellow, time in ms. Gotcha. I'm wondering which direction the arm was going on the test, lol. I agree with sata over scsi as well.

[Scott Gold]

I know this is an old thread but I have a question regarding SATA 1.5 or 3.0.

I'm leaning towards 10,000rpm SATA drives but I'm noticing that I'm having a hard time trying to find a 3.0 one.

Do they even make 3.0 10,000rpm SATA drives?
If so where?

What would you recommend, 7,200 3.0 SATA or 10,000 1.5 SATA.

Thanks,

Scott

[Daniel Browning]

Originally Posted by Scott Gold

SATA 1.5 or 3.0?

It doesn't matter. 1.5 has enough bandwidth for 192 MB/second, which is easily 50% more than any disk on the market can sustain.

Furthermore, spindle speed (7K, etc.) isn't the only performance indicator. My Samsung Spinpoint F1 handily bests my Raptor 150 10K by a wide margin in sustained throughput, but not latency. So the only way to really know is performance testing.

Also remember to pay attention to the entire storage chain: disk (10K), connector (SATA), controller (3ware), bus (PCI Express X8), and driver (stable and optimized?). It's only as strong (or fast) as its weakest link, and there are a lot of hidden gotchas.

Remember too that you can trade capacity for speed; the faster, outer portion of a bigger disk may be the same capacity but an even higher throughput than a smaller, higher rotation disk.

[Andy Mace]

All the 10k drives I've seen are version 1.5. I don't think it matters. having lost a pile of Sata drives over the past few years (I work on computers) to what I suspect is mainly heat, I'd recommend going with something that runs as cool as possible. The information over at storagereview.com might be of some help.

[Don Blish]

I ran SCSI drives for years...and in my StdDefinition editing PC (1999-2005). As a retired IT manager, there is no longer any reason to use the SCSI interface. Quality underlying drives are the same. SATAII's 3gbits per second is equivalent to SCSI 320m Bytes per second. The last advantage SCSI had used to be called "scatter/gather" and is equivalent to SATA's "command queuing".

[Mike McCarthy]

Interesting that SAS was not mentioned. That is one reason why we haven't seen 3Ghz SATAII drives at 10K RPM yet, that and because WD is the only manufacturer. If you want a high end, high speed, low latency, disk array, SAS will be better than SATA, and way, WAY better than SCSI. For most any other scenario, the lower price of SATA disks overcomes any potential advantages of SCSI or SAS, by allowing one to use more spindles for the same price, and allowing much larger total capacities for a given price.

For most video editing work, go SATA for sure. DI CCR on uncompressed DPX sequences, you may benefit from 15k SAS drives. Database servers, you may benefit from the lower latency and access time on SAS. The other 99% of you, SATA is the way to go.