RAID Edition 5 ("RE5") series of rotating hard drives

[http://forums.storagereview.com/index.php/topic/29262-ssd-reliability/page\_\_st\_\_30\_\_gopid\_\_271041#entry271041](http://forums.storagereview.com/index.php/topic/29262-ssd-reliability/page st 30 gopid 271041#entry271041)

I’m watching the SSD marketplace very closely, and everything I read above still convinces me to stay with HDDs, until the SSD world improves a LOT!

Now hear this …

Consider a RAID Edition 5 (“RE5”) line of WDC HDDs.

Here’s why:  I can’t find a 6G WD HDD with 64MB cache that also supports their Time-Limited Error Recovery (“TLER”) which they highly recommend for all RAID arrays, unless I buy their expensive SAS HDDs, or spend a lot on Velociraptors.

For example, at the WD website they warn against using multiple Caviar Blacks in a RAID array, because those HDDs don’t support TLER.  The other reason is that, generally, higher performance comes with higher cost and with higher capacity too.

So, imagine a WD RE5 HDD with 64MB or 128MB cache, 6G interface, 7,200 rpm, and perpendicular magnetic recording (“PMR”), but in relatively small capacities e.g. 64GB, 128GB and 250GB – targeting the SOHO market sector (initially).

[“SOHO” = Small Office / Home Office]

With the right pricing, 4 x HDD caches @ 64MB = 256MB in RAID 0 (HDD caches are effectively additive in RAID 0 mode).

The closest I can come to this goal is the WD1500HLHX:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822136928&Tpk=N82E16822136928

… BUT the cache is too small, and for that price ($120) the capacity is too large for what I am suggesting here (or, for that capacity the price is too high for what I am suggesting here).

For comparisons, we have measured 283 MB/second with 2 x WD5003ABYX in RAID 0, using the ATTO benchmark:
 that HDD has a 64MB cache, 7,200 rpm, PMR, but only a 3G interface:

http://www.supremelaw.org/systems/io.tests/WD5003ABYX.2xRAID0.jpg

I would jump tomorrow for an inexpensive 64 or 128 GB version with PMR, 7,200 rpm, and either a 64MB or 128MB cache, as long as its interface were also the 6G standard (600 MB/second).

This “RE5” series would be designed specifically for RAID arrays where TLER is recommended, and where the combined HDD cache total creates a large effective hardware buffer that helps to eliminate the need for a dedicated I/O buffer on-board expensive RAID controllers.

Now, let’s look at some realistic prices: the WD2503ABYX retails for only $70 at Newegg:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822136797&Tpk=N82E16822136797

Let’s reduce the capacity to 64GB or 128GB, keep the cache at 64MB, keep TLER, and up the interface to 6G:
WD should be able to sell such a “RE5” HDD at an MSRP at or near $75 USD.

So, 2 of these in RAID 0 cost ~$150 and should exceed 290 MB/second sustained.  4 of these in RAID 0 cost ~$300 and should exceed 500 MB/second sustained.

Our experimentation here was able to reach 618.5 MB/second, but that speed required 4 x Hitachi 15,000 rpm 2.5" SAS HDDs at $250 each ($1,000 just for HDDs and not counting the additional cost of a Highpoint 2720 controller):

http://www.supremelaw.org/systems/hitachi/Hitachi.C15K147.ATTO.4xRAID0.JPG

Summary:  Until SSD technology matures a lot more, RAID arrays of relatively small but relatively fast rotating HDDs can perform quite admirably for much less money and with much greater reliability than current SSDs are able to demonstrate.  This kind of product would be very appealing to the SOHO marketplace who are understandably wary about paying high prices for relatively poor reliability.

P.S. Just one man’s opinion here: your responses are all welcome!

MRFS

A little follow-up:

I just completed a test with the following system, stock settings in all cases:

ASRock G41M-S3 motherboard

Intel 3.2 GHz Pentium 4 model 640

2 x 2GB Corsair XMS3 DDR3-1333 memory (4GB total, dual channel)

used Highpoint RocketRAID 2322 PCI-E x4 RAID controller w/ original driver

2 x brand new WD2003ABYX HDDs in RAID 0 (purchased from Newegg)

ATTO Disk Benchmark v2.46

Here’s a screen shot of the ATTO results:

http://www.supremelaw.org/systems/wd/2xWD2503ABYX.RAID-0.RR2322.G41M-S3.JPG

Max observed READ speed is 296.2 MB per second.

This rate is just a fraction short of the SATA-II bandwidth of 300 MB/second

i.e. this READ speed is slightly better than what is being observed with

the best single 3G SSD, regardless of vendors.

Total cost of the 2 x WD2503ABYX at Newegg:  2 x $70 = $140

(PCI-E controller for either the x1 slot or x16 slot is a separate cost item)

Bear in mind, we used the only x16 PCI-E slot for this experiment:

for this type of machine, we would be more than happy to

upgrade the on-board video with any of a number of excellent

x1 PCI-E video cards that are now available, from different vendors.

Because the ASRock G41M-S3 also has 2 x legacy PCI slots, one of those

can be easily assigned to a Gigabit Network card:  32 bits @ 33 MHz

= 1,056 Mbps. which is just enough bandwidth for a single Gigabit NIC.

We did this experiment because the total cost of this DIY system

is very reasonable, particularly for SOHO settings which need

a reliable backup storage server, preferably with a Gigabit LAN.

Our next experiment will be to upgrade the RAID controller to

a Highpoint RocketRAID 2720SGL, and install Windows 7

to the same 2 x WD2503ABYX HDDs wired to that 2720SGL.

We’re hoping it will exceed the 300 MB/second threshold

with the faster controller and more recent driver for Windows 7.

Stay tuned:  we should have more to report in about 2 weeks :slight_smile:

MRFS

We break the 300 MBps barrier with the latest driver for the RocketRAID 2322 –

307.1 MBps:

http://www.supremelaw.org/systems/wd/2xWD2503ABYX.RAID-0.RR2322.G41M-S3.latest.driver.JPG

As such, we’ve exceeded MAX HEADROOM for the 3G SATA-II bandwidth of 300 MB per second.

MRFS

Our colleague at www.pcper.com – Allyn Malventano – asked the following question:

>  Are you proposing an independent head RAID-within-a-drive?

Our answer follows:

No … just a simple & low-cost RAID 0 that can exceed 300 MB per second using just 2 x rotating HDDs, and each HDD has a relatively large hardware cache, even though each rotates at only 7,200 rpm.

The WD2503ABYX also has a 5-year factory warranty and this experiment configures 2 of those in a single RAID 0, with XP/Pro SP3 installed in a 30GB “short-stroked” system partition.

(Compare 3- and 5-year warranties by dividing retail cost by warranty years:
the 5-year “enterprise” warranties usually come out ahead on this measure.)

All ATTO measurements are being done on that short-stroked C: partition.

The ASRock G41M-S3 motherboard is another way to save money:
the single x16 slot is only Gen1, but it still has plenty of bandwidth for this experiment.

According to the specs I’ve found on the Internet, that motherboard has a Gen1 x16 slot and a Gen2 x1 slot: this is a feature of the G41 chipset.

(If upgraded video is required, there is that Gen2 x1 slot for compatible video cards:
Newegg has quite a few x1 video cards, some with support for 2 monitors;
nevertheless, for applications such as an inexpensive storage server,
the integrated graphics are just fine; and there are 2 x PCI slots,
one of which can host a Gigabit NIC.)

And, DDR3 memory is also very inexpensive now: 
this machine has a matched pair of Corsair XMS3 DDR3-1333.

I think I paid only $120 for that motherboard + memory + one LG IDE optical drive.

Today, I updated the driver in my used Highpoint RocketRAID 2322:
http://www.newegg.com/Product/Product.aspx?Item=N82E16816115028
and was able to break through 300 MBps here with those 2 x WD2503ABYX:

http://www.supremelaw.org/systems/wd/2xWD2503ABYX.RAID-0.RR2322.G41M-S3.latest.driver.JPG 
(307.1 MBps)

Next step is to replace that older controller with a slightly more expensive RocketRAID 2720SGL
which is on order and should arrive in about one week:

http://www.newegg.com/Product/Product.aspx?Item=N82E16816115100&Tpk=N82E16816115100

(I’d also like to see Highpoint offer the RR2710 in a “SGL” variant:
http://www.newegg.com/Product/Product.aspx?Item=N82E16816115084&Tpk=N82E16816115084
because only 4 such 6G SATA/SAS channels are all that may be needed for the SOHO settings I’m contemplating i.e. much higher densities per HDD, soon to exceed 3TB per drive using Hitachi’s latest announcement of 1TB per platter)

I know you do prefer much better RAID cards, but the experiment here is to deliver decent performance on a very restrained budget e.g. typical of SOHO settings – in recognition of nationwide economic conditions.

I’ll measure the effect of this upgraded controller, then I’ll upgrade the OS to Windows 7 Ultimate 64-bit version and repeat the same measurement.

Finally, funds permitting, I can also upgrade that G41M-S3 motherboard to 1333 MHz front-side bus and DDR3-1333, for a 1:2 FSB : DRAM ratio (333:667);  this change must be done with jumpers instead of BIOS settings.  But, I’m expecting minimal effects from faster processors, because the Pentium 4 model 640 is already 3.2 GHz, and these experiments are I/O-bound.

One key objective of this experiment is to substantiate my proposal to Western Digital that they upgrade their 7,200 rpm RAID Edition HDDs with 6G support + relatively smaller capacities but with relatively larger caches e.g. 64 or even 128MB per HDD.  In this way, a RAID 0 array effectively adds the caches for all such RAID members, and this “cumulative cache” can help to eliminate the need for a large hardware cache on-board the RAID card e.g. on Areca cards – further reducing overall cost.

If you start reading this thread from the top-down, that proposal will make more sense.

I think there is no argument that enterprise-class storage remains very expensive e.g. 6G Velociraptors or 15,000 Hitachi 2.5" SAS HDDs, and thus remain mostly out of the reach of SOHO budgets:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822145462&Tpk=N82E16822145462
http://www.newegg.com/Product/Product.aspx?Item=N82E16822145319&Tpk=N82E16822145319

I’d like to bring more powerful storage to SOHO settings, without the premiums which enterprises are willing to pay and SOHO budgets cannot afford.

Lastly, there is still too much uncertainty associated with SSDs, if my reading of recent industry reviews is representative of the current installed base of SSDs.

Many thanks for your interest, Allyn.

Cc: Ryan Shrout at www.pcper.com , and also hardocp.com


Sincerely yours,
/s/ Paul A. Mitchell, B.A., M.S., Instructor,
Inventor and Systems Development Consultant

All Rights Reserved without Prejudice

We recently received a brand new Highpoint RocketRAID 2720SGL

http://www.newegg.com/Product/Product.aspx?Item=N82E16816115100&Tpk=N82E16816115100

and installed Windows 7 Ultimate 64-bit version on the same

RAID 0 array, with 2 x WD2503ABYX HDDs formatted with a

“short-stroked” C: system partition of 30GB.

Here’s the ATTO Benchmark result for C:

http://www.supremelaw.org/systems/wd/2xWD2503ABYX.RAID-0.RR2720SGL.G41M-S3.Win7.x64.JPG

i.e. very close to the result we obtained with XP/Pro SP3 and the

Highpoint RocketRAID 2322 (see above).

For those who are interested in evaluating the 2720SGL,

this is a stripped down version of the RR2720, but

the retail package does NOT include any manual, cable(s)

or software disc:  drivers etc. must be downloaded first

from the Highpoint website and then written to some

medium that can be accessed by the Windows 7 Setup disc.

Highpoint’s README.TXT file explains how to load the driver

in various situations e.g. fresh OS install, or existing system:

we’ve archived that README.TXT file here:

http://www.supremelaw.org/systems/highpoint/2720SGL.Readme.txt

Also, one or two “fan-out” cables must be purchased separately,

because the 2720SGL does not include cables either.

Here is the Highpoint cable that works with this controller:

http://www.newegg.com/Product/Product.aspx?Item=N82E16816115064&Tpk=N82E16816115064

Some people have understandably complained about the long 1 meter length

of this fan-out cable;  however, identical 0.5 meter cables are available from

other vendors cf. “SFF-8087” aka “Internal Mini-SAS to SATA Cable”.

To ensure proper connections with that controller, we find it is

easier to connect the multi-lane end of the cable BEFORE installing

the controller in an available PCI-E slot:  in this way, the connector

can be visually inspected to confirm that it is properly seated.

Other than that, the fan-out connectors are standard SATA style,

and we also have that same cable working fine in another workstation

with 4 x Hitachi 2.5" SAS HDDs.

Sincerely yours,
/s/ Paul A. Mitchell, B.A., M.S., Instructor,
Inventor and Systems Development Consultant

All Rights Reserved without Prejudice