Help with new build using older OCZ SATA3 SSD's

erpsaa

Honorable
May 17, 2012
46
0
10,540
Hi,

I'm hoping to get some help and advice on a new build regarding storage.

My last build (almost 4 years old now) I completed with the core system components being an ASUS LGA 2011 P9X79-E WS, the Intel Core i7 3960x, and the LSI MegaRAID 9271 8i PCIe3 Controller. (SAS to SATA600)

The storage I used are now the legacy OCZ Vertex 3 SATA3 240GB SSDs. I initially purchased 9 x 240GB drives (1 extra) and am still using the same original 8 configured as RAID-0 on the same LSI controller. Not one has gone down once in almost 4 years. The virtual drive is about 1.8TB. I have about 800GB of apps and data on this WS.

The few times I've had crashes over the years which needed to be restored, the LSI Controller, the Legacy SSD’s and the PIT Snapshots I’ve set EMC NetWorker to create about every 8 hours, (or on-demand) takes about 15 minutes in TOTAL including a reformat the 1.8TB RAID-0 array followed by a restore the most recent, or whichever PIT snapshot I decide to use.
That said, the new build I've decided on (core components) will be:

The ASUS X99-E Workstation / USB 3.1 (LGA 2011 V-3) Mainboard and the Broadwell-e, Intel Core i7 6900. (So far, most every bench and review I've read seems to indicate spending the extra $$$ on the 10 core 6950x vs the 8 core 6900 is a waste of money). I welcome any opinions on that as well.

Given the new build, can anyone tell me if I'm making a mistake using the same storage, at least to start out? The newer LSI Controllers when matched up to my current 9271 8i have not evolved that much to persuade me to change, not unless I were to move to LSI's 12 GB/s SAS HBA's or Controllers which is not necessary for a Workstation I use for work, but from my home office.

Lastly, I can ask this under GPU questions, but thought I'd add on a quick GPU question here as well since it's related.

I am currently using 2 x 27" monitors with a single EVGA GTX 680 4GB Memory and that works fine for the type of Apps I run (I hope).

Mostly, ERP, Adobe Design and Web Premium CS6, and several other apps, including one that indexes a very large number of electronic books and media resources, all of which work with each other and perform active web sync. I also run VMware WS version 12.5 for Linux (Debian).

I was always of the belief that the fastest storage arrays combined with a CPU that can be OC'd and remain stable were all I needed for max performance.

However, I don't run any games and I've come to learn that people who do run games, fill in knowledge gaps I have, despite being in IT for many years and accustomed to selling and building servers and Business Workstations. Thanks for any help.

Dean
https://www.erpsaa.com/

.
 
The storage I used are now the legacy OCZ Vertex 3 SATA3 240GB SSDs. I initially purchased 9 x 240GB drives (1 extra) and am still using the same original 8 configured as RAID-0 on the same LSI controller.

I have a question on this:
With this SSD RAID 0 array, is there any actual, perceived, real world performance benefit over a single large SSD?
And what do you run on this?

Many people come here looking for a RAID 0 + SSD solution, and many of the tests out there show no actual benefit outside of a few limited use cases.

Your 8 drive array is out on the edge of extreme, so I'm highly interested in how it actually works.
 


Sorry for the delay in reply. I checked off to be alerted if anyone posted and no notifications came through. I'm not sure why email alerts are not working, that never happened before. That said:

First and most important USAFRet... I’d like to say I’m very thankful for, and appreciate your service. Let me do my best to answer your questions as quick and direct as possible. You asked:

"With this SSD RAID 0 array, is there any actual, perceived, real world performance benefit over a single large SSD? And what do you run on this?"

In short, there is much more than a performance benefit, however, performance tops the list. Before I get to the actual performance, here is how performance is achieved with a HW RAID Controller and I’ll explain why with an analogy. A few Laptops these days offer two drive ports enabling one to install say 2 x 1TB SSD's. One problem is these Laptops are overpriced, but a second and more critical problem is these Laptops use BIOS/Software RAID. A PCIe3 “ROC” Raid on Chip Controller (with usually 1GB of DDR3 RAM) uses its own processor and memory. This offloads CPU and Memory utilization and the Controller Memory provides on-controller caching, thus speeding up access times by using the (off the host) DDR3 Memory for cache while also permitting something you cannot do with BIOS RAID, Write-Back Caching. The I/O is using the controller’s processor, allowing multiple disk arrays, with no effect on the Host CPU.

HW RAID Controllers all have a separate and distinct BIOS which is eons ahead of what someone will see at bootup they press CTRL + I (to launch the Intel IRST Software RAID Controller) or CTR: + M (The Marvel Software RAID Controller) just two examples, both of which are part of the Mainboard’s BIOS. This is key, since a HW RAID Controller is independent of the host at Bootup, security is enhanced, as well as boot errors minimized.

The final and most important part of using a HW Controller is backup and restore is about 10x as fast. With all the SATA ports on my LGA 2011 and the more with the x99 Chipset on my new LGA 2011 V-3, I will still only need to use 2 ports. Those are for the 2 optical drives I like to have.

Assuming a HW Controller has 8 x 250GB SSD drives at RAID-0 (each of which perform at say 550 GB/s Read and 520 GB//s Write, and if a quality SSD, say 90,000 IOPS each) in the end it’s a 2TB (Logical or Virtual) C:\ drive. All you need for easy back up and restore (I’d don’t mean file backup, I mean the OS, Applications and files) to a single equal size drive you can boot with if the array crashes, is a minimum of one (but I suggest 3) SINGLE 2TB HDD’s.

Backing up RAID-0 is backed up the same way a single drive is backed up if the Application a person is using does what people today call Cloning and what’s been called for years in IT departments “A Bootable Image backup.” The term is still used today. A more familiar term I think is Point-in-Time snapshots (PIT). PIT’s are created at predetermined times using Replication or deduplication, and if an array or drive gets corrupted, the exact time of the corruption is determined and a restore is done by selecting one of the PIT’s.

The issue with RAID-0 has never been performance, (outside of Enterprises) it’s been an issue of backup and restore. The number of times I’ve heard people say: “Never use RAID-0, there is no redundancy if you crash you lose everything.” That statement has always puzzled me for two reasons.

1. Has anyone ever heard of an Enterprise that does not use a RAID array in their servers? I don’t mean someone who has built a home server, I mean an Enterprise Server Farm. Why do they all use some type of RAID at all, why not a large single drive? (In an enterprise server, today; the most common server array is a 4-20 drive RAID-10 (or RAID-1+0 array) which I’ll explain below.

2. If someone is using a single drive, what are they using for data protection? Or, backup and restore? Many people today still say they “use a backup program to back up their files to an external drive.” I’m tempted to ask them, if you save your files in an organized manner, why are you using an Application? Just drag them over to the external drive every few days.

Yet, many people have discovered there are a number of applications which Clone their single C:\ drive, so if it gets corrupted, or dies, not only the person boot up with their Cloned backup, (not having to reinstall Windows, all their Applications and files) the same software is on the backup, so all that is needed, is to after booting up with the Cloned backup (if the C:\ drive issue was just corruption and it did not die) reformat the C:\ drive and restore it with the backup, they boot up like nothing happened. If the drive died, you have the backup to boot up with and use until you but a new drive. (That’s why I suggest having three bootable backup drives).

That brings us to the question: What’s the difference if you were to back up and restore a C:\ drive created using 8 x 250GB SSD’s for a total of 2TB vs. backup and restore of a single 2TB C:\ drive as described above, with a single bootable backup? NOTHING. No difference at all. Windows XP, 7, 10, Server, Linux, etc. all see the RAID array as a single drive. If someone knows how to backup and restore a single drive with good cloning software, RAID-0 with 20 100GB SSD’s is no different.

So, why would Enterprises all use RAID-0 in their Servers and in their Workstations (the ones using high end applications) like DNA modeling for example?

Also, what about the new M.2 PCIe SSD’s? Why are they so fast and what’s the raw difference in speed between SATA3 and PCIe3? 6 GB/s or 750 MG/s vs 985 MB/s. Multiply by x4 and you have the highest end M.2 PCIe3 x4 SSD, which moves about 3GB/s Read and 2GB/s Write. A nice step up, but they are a consumer product and I’ll only mention one thing that lack, the ability to perform cache write back, which is why the delta between the write and read speeds is such a gap.

Finally, the very simple explanation on RAID-0 performance and why RAID-10 is run on most servers. Again, this is all about math.

Take one SSD that runs at 550 GB/s Read and 520 GB/s write with 90,000 IOPS. A PCIe3 RAID Controller allows for almost perfect scaling. 8 x 550 MG/s = 4.4 GB/s, 8 x 520 GB/s = 4.1 GB/s and 720,000 IOPS. A RAID 0 array is striped, which allows the user to select the stripe size based on the type of Applications they are running and the work they are doing. 128 is a typical stripe size. To put it another way, data is written and read simultaneously using 8 SSD’s each of which is moving at 550 MB/s R 520 MB/s R and the data is broken up into chunks. So, even a two drive RAID-0 array is double the speed of a single drive. The reason I would not do it, is because you’re limited to software RAID with a Laptop. To build a Workstation with a controller with only two drives at RAID-0 with the cost of SSD’s today does not make sense, but you’ll be moving faster than 1.

Want to move even faster? 12 GB/s? Then you need to go with SAS Controllers and SAS drives ($$$$$$) expensive, moving at 15,000 RPM’s. They are not limited by the SATA3 highest speed per drive, but can use the full bandwidth of PCIe3 985 MB/s and also go up to PCIe3 x8. X16 is not here yet with PCIe3, only with PCIe2, which caps out at 2.7GB/s.

That’s why if you do a search on Hardware RAID controllers, what you will find are models with 4 internal, 8 internal and 16 internal.

Why do most Enterprise Servers run RAID 10 these days, with 16 drives? This accomplishes an 8 drive RAID-0 array for the max SATA3 speed, matched with 8 mirror drives, so if one drive goes down in the RAID-0 array, the 8 drive RAID-1 mirror takes over and becomes the new 8 drive RAID-0 array. That is used for fast Failover, whereas in the past, or if a business wants to save money, they build RAID-5 or RAID-6. RAID-5 using a minimum of 3 drives. One is not used except to store backup Parity of the other two drives, which are running at, guess what? RAID-0? RAID-6 has two redundant drives. But, they are not good for Failover. Failover needs to be FAST, NO DOWN TIME. That’s the reason for RAID-10.

NOTE: They are still working on RAID arrays which need to be backed up, which is where disaster recovery comes into play.

This is one of the most common Server Cards sold today:

https://www.broadcom.com/products/storage/raid-controllers/megaraid-sas-9361-16i

Here is a decent article from PC World in RAID-0.

http://www.pcworld.com/article/2365767/feed-your-greed-for-speed-by-installing-ssds-in-raid-0.html

Finally, you asked what I am running. Tier I ERP, Oracle (Financials, Procurement and BI Analytics). Adobe Design & Web Premium CS6 (Mostly Dreamweaver and Photoshop), X1 Enterprise Search, Logos.com (Application that manages a large indexed Library, over 9K+ books, about 200GB for the index, which works also with live web sync), VMware Workstation 12.5/Linux, EMC NetWorker v.9 aand normal stuff, Windows 365 Business, etc.

Hope this helps.

Dean
https://www.erpsaa.com/
 
Right.
The OS sees a single drive, or a RAID 0 array as a single volume, and would be accessed and backed up the same way.
And we all know RAID 0 is not a backup. People see the word RAID and assume...

I'm just curious about the actual (verified) performance difference between n drives, or a single drive of the same aggregate size.
In an enterprise situation, where you have a LOT of data throughput...sure.
Otherwise, not so much.
 
"I'm just curious about the actual (verified) performance difference between n drives, or a single drive of the same aggregate size.
In an enterprise situation, where you have a LOT of data throughput...sure.
Otherwise, not so much.[/quotemsg]"

Absolute agreement on that.What is interesting is that the two most critical functions are running VMware WS 12.5 (Linux Debian for Remote Pen Testing and I forgot to mention, vCloud Air, where in emergencies I sometimes need to deploy DaaS). That Logos.com app is the worst. It's very slow on my Laptop and the 1TB Samsung 850 Pro I use is an excellent SSD. 13 seconds just to load. It's almost instantaneous on the Workstation.

Cheers,

Dean
 

TRENDING THREADS