Next generation Sata Drive comparaison

[BB_One]

Excerpt:

The last time we did a HDD review here at HCW was in 2003 where Ed and Carl took a look at what was at the time, the fastest HDs around. Things have certainly changed since then with the even larger disk sizes as well as almost wide-spread adoption of SATA drives. We're even beginning to see SATA II drives on the market that offer an interface speed of 3.0Gb/sec as compared to the 1.5Gb/sec offered by SATA. Of course, we don't expect to see real world performance close to that, but we can certainly hope!

Comparaison is here (http://www.hardcoreware.net/reviews/review-277-1.htm)

Special note from BB_One:

What gets to me in this or similar comparaison when discussing SATA II, is the fact that the reviewer omits to consider that although Native Command Queing is a SATA II specifications, so is a burts transfer rate of 3.0 Gb as oppose to 1.5 G for Sata I. In this specific comparaison, the reviewer identifies the Hitachi T2k250 ( my next data storage drives), as the best performing one after a Raptor. Off course it will be, it is the only one supporting 3.0 Gb burts transfer rate, the other 2 do not! Hence and sadly the reviewer fails to highlight this issue. The Maxtor or Seagate are partially Sata II compliant. Do note that your chipset ( hard drive controller has to support Sata II to enjoy it's benefits)

[FB]

Seagate are partially Sata II compliant. Do note that your chipset ( hard drive controller has to support Sata II to enjoy it's benefits)

thats me :( i have 2 SATA drives with NCQ, but not SATA mobo features.

[BB_One]

Arggghh... darn edit function

The Rapor off course is not Sata II compliant and is there for comparaison purposes, also note that it utilizes Tag Command Queing as oppose to Native Command Queuing. To this effect, some testing demonstrate that TAG can be detrimental at times under windows, be aware of this for the controller is looking for NCQ not TAG.

[Hawk]

Hopefully the current chipsets work with these with a simple BIOS upgrade. . .

[timpanogos]

dumb guy question.

1. My 10k rapture is going to run stand-alone on my nf4, right?? (please say yes)
2. I read that running 2 drives in RAID actually does not make sense because it actually doubles your chance of fault, instead of increasing your chance of recovery (i.e like with 3 or more drives in RAID). So why are the double raids so popular .. since they are stripped they output faster??

thanks,, and sorry about all my traffic

Chad

[RyderOCZ]

That is correct Chad...2 drives in raid0 (striped) can move data 33% faster than a single drive can. 3 drives are faster, 4 even more, etc.

Raid1 (mirroring) makes each drive an exact duplicate of each other, therefore providing security if 1 drive fails.

Raid5 (data striping with Parity) has similar benefits of Raid0 for speed, but with the redundancy of Raid1.....3 drives or more are required for this type and most embedded motherboard raid controllers do not offer this. I can explain this more if you like :)

[timpanogos]

wow, 33% is well, 33%, big increase! I've just been at back end servers too long, where only concern was hot-swappable drive failure RAIDs. Did not realize that these also gave a nice speed bump also! cool. (I've never set them up or delt with them personally, just allways knew our Oracle DB's were always running on those RAIDs).

Dang, we have some DB's that are smaller that we really could use the speed on .. After reading BB's article link and this info, need to look at 4 raptors and see what they's do!

So will we expect to see SATAII in larger drives at 10k anytime in the near future?

Chad

P.S. please tell me my nF4 is going to come up on 1 SATAI (raptor)

[RyderOCZ]

ooops...sorry for that chad....yes 1 drive will work just fine....you will not need to install any Raid drivers...and make sure not to set the Raid function on the Sata controller to enabled, XP will recognize the sata drive just fine :)

[Snafu]

So will NCQ actually be a benefit to anyone who does not run HDD intesive apps and with multiple apps running? Sounds like it is more a benefit for servers than for you and I.

[RyderOCZ]

NCQ will help all SATA drives just like it speeds up SCSI drives. The controller and the Drive must support it though.

NCQ sorts the requests before they are sent to the drive based on where the read/write heads are, so the drive is not wasting time going from one extreme to the other if there is a request for data in the same area that is coming in like 2 cycles.

[timpanogos]

Thanks RyderOCZ!! (bloodpressure coming back down - but eye now on that other 33%)

So, I'm a bit surpised that that 2x in bandwidth (1.5g to 3g) on the Hitachi T2k250 did not at least catch up with the raptors 25% advantage in platter speed?.

So BB why the hitachi over the raptor when the raptor is still significantly faster?

[Snafu]

Not arguing that it does sound like it has an advantage but if you do not run multiple apps that are HDD intensive and your properly maintain your dirve (i.e., defrag) then would you or I outside of the server setting see an actual advantage. I wonder if there is a good bench to check this?

BTW didn't Hitachi buy IBM's deathstars?

[timpanogos]

Did you really want to know???

Well, as I programmer (multi-threaded programmer) .. I can say this, about that ... a HDD intensive program is actually going to be one of two flavors:

1. serial io intensive
2. random io intensive

All applications will actually end up serial io, but how much they jump around and how big is that ultimate serial stream is the question. (BTW this is a similar problem for TCP/IP streaming also) Many small bursts, verses fewer large bursts is always the battle (PPS verses bandwidth) High PPS always equates to higher CPU load (more packet processing) and of course the latency per packet reduces your throughput potential.

I worked at Novell for 10 years. Early on (SPX/IPX days) we wrote what was then called "PacketBurst" which is now a very common pattern for handling just this. How to pack lots of PPS in a single large stream, for not only lower CPU utilization but much higher throughput (approach bandwidth as close as you can).

So what does this all have to do with anything?

If you are not a database, spreadsheet or some type of application that extends random access from a caching scheme to persistence (disk), then you are likely a simple file streaming type of application (loading executables, delivering web pages, loading word docs ... etc ). In the other cases, you are looking/using the disk as simply an extension for memory that you utilize some form of random accesses on (btrieves etc.)

Of course the random access types of applications are going to GREATLY benefit from the NCQ stuff ... but what about that file streaming?

In these cases, of course disk-defrag is going to help you as it brings the file all together on the disk. The NCQ could actually eliminate the advantage of the defrag … if it was not for the head movement time loss … same concept/pattern as packet burst – pack many small ones in one large one. So of course defrag is still good stuff. But is that the end?

See the trick is, with the NCQ, that not only does it save on the seek times, but it brings the smaller packets together for potentially larger bandwidth bursts to the buss.

Now I have no idea how they have actually written these NCQ’s, and the on-disk caching is going to couple very closely to the vendors NCQ schemes … but, if a given application was accessing MANY small files (ok, back to web server, sorry) then the ordering of these many files “file-bursting” them could be a huge boon.

So, does any of this have worth for the workstation? Age old – that depends. It is not necessary to have multi-threaded (or multiple applications) accessing the Disk to see an advantage. It’s based on – am I random access file type of application, yes, I’m going to benefit from this greatly. Am I a file streaming application that accesses many files (i.e. directory/file scanning of some sort), yes, I’m going to benefit from this (if the NCQ implementation/on-board cache tries to do some sort of file-burst – i.e. several files on buss at once)

Now I’m not a gammer by any means … but it would not surprise me if most games are not architected as a huge memory model, where the idea is to cache (in memory) as much of the game database as possible, but update the cache from disk as a player moves out of the cached model. In otherwords, games who’s databases are on disk, maybe VERY Random accessish in there architecture and may actually benefit from the type 1 disk accessor mentioned above.

Bottom line … who knows (but is all sounds good eh)!

Chad

[RyderOCZ]

:eek: :look: :eek:
Ummmm yea what he said, LOL

Good Stuff there Chad, makes an excellent read :)

[Snafu]

:yikes:

Okay who was the wise guy and asked the question :scratch:

Definitely good stuff in there. It almost reminds me of the question whether raid 0 is worth it for the home user (no, no, no - not asking btw :lol: )

I suspect some of us at BE play a game or two and would be interested in know if NCQ woudl help the gamer is either loading the game or loading a level. Sounds like it will depend on how the game is loaded onto the system.

Cheers mate

[BB_One]

So BB why the hitachi over the raptor when the raptor is still significantly faster?

Simple answer, I stated Hitachi would be my data drives ( only waiting for Hitachi to release their 250 G drives, which after 5 months of waiting are still not on the market :scratch: , I already have 2 x 74 Gigs Raptors in Raid 0 as main boot drive and speed dependent applications/games. :lol:

PS: U may be asking a lot of questions, but your experience sure pays off and your understanding of the architecture certainly bang on !

Snafu:

Yes Hitachi bought the Hard Drive division of IBM. Personally I have had a few Deskstar from IBM, fast , reliable and never had any problems. Sadly I am not hearing this from everyone. However, I looked long and hard before making my decision and Hitachi is the way I am going on this one. Sadly Seagate ( would have been my prefered choice, as I have been very please with the SCSI drives that I have), but no full SATA II compliant drive on horizon.

[BB_One]

So, I'm a bit surpised that that 2x in bandwidth (1.5g to 3g) on the Hitachi T2k250 did not at least catch up with the raptors 25% advantage in platter speed?.



Simply because it is not true 2 x , do note the terms used: BURST transfer rate. Burst being a keyword here, for as you so nicely explained in you previous post. It depends of the type of data being accessed. Small chunks or large chunks? kind of reaches what Snafu was raising. Sustain rate or burst rate. In real life numbers, again depending on the data being retrieved you may or may not see major improuvement. Here is a real number for day to day use:

Raptors:

Buffer To Host (Serial ATA) 1,500 Mbits/s (Max) (Sata I)
Buffer To Disk 72 MB/s (Sustained) >This is max off course, nominal not stated!
Read Seek Time (Average) 4.5 ms SCSI legacy with a 10000rpm !

Hitachi T7K250 (Sata II)
Interface transfer rate (max. MB/sec) 300 ( hence Sata II)
Sustained data rate (MB/sec) 67.8- 32.9 (zone 0 - 29) <--- Buffer to disk
Seek time (read, typical) 8.5ms


In short think of the Sata and Sata II, as you would with Parallel ATA Drives, in short

P- ATA = 133 MB
Sata = 150 MB
Sata II = 300 MB.

Sata III about to be finalized/ratified in the approval clouds and Sata VI already claiming to deliver a theoritical 2 GIG!

Interesting times ahead, soon one will wonder why go SCSI !