RAID 0 SSD Array Slow...

[Makaveli]

Let me put it this way: the OP has 2 Vertex 3s on an AMD motherboard, so none of that BS matters one little bit.

Now, has the OP benched the drives again with read-ahead and write-back caching?

I wouldn't consider it bs its a fact. However since the OP is talking about a Homebuild the comparison is irrelevant to the topic at hand.

Alot of questions have been asked so everyone is waiting for feedback from the OP. This thread is an interest to me because i'm going Raid 0 shortly so always good to keep an eye out for issues even if were on different setups.

 

[Cerb]

I wouldn't consider it bs its a fact. However since the OP is talking about a Homebuild the comparison is irrelevant to the topic at hand.

That the OP's SSDs are, "junk," is not a fact, the OP has made no mention of BSODs, which are hardly ever an issue nowadays. The only facts were RMA rates, and that the performance would be lower than more expensive SSDs (and that's part of why those SSDs could command higher prices and fewer deep sales in the first place!), and a little slower than an Intel drive controller. Both of these are things the OP aught to have known, but both of them also should only account for 10-20%, not 50%.

Most importantly, the RAID is not improving sequential performance, which should have zero to do with the performance/quality of the individual drive, and the OP already has the hardware, and has no apparent problems with it, except for the lack of improved performance in RAID, so far as we know.

I wouldn't get one of those OCZ drives, either, but for the OP's issue, it is pointless BS, and little more than thread-crapping. He should leave that to hot deals threads on OCZ drives, and SSD recommendation threads.

I am interested to know if write-back caching being off by default was the main culprit (read-ahead can affect simple benchmarks and real workloads very differently, sometimes, so I can understand it being off by default, even if it were to lead to lower sequential read benchmarks).

 

[sequoia464]

Now, has the OP benched the drives again with read-ahead and write-back caching?

I just checked and the last bios Gigabyte put out for the OP's board is from December of last year, unless the OP can find a modified bios I doubt that he will be able to enable read-ahead and write-back.

Even though those options are shown in RaidXpert - they can't always be enabled. I was running a modified bios on my Sabertooth 990FX until Asus finally included the new raid rom in the stock bios a couple of months ago.

The OP should still enable NCQ through RaidXpert as it isn't enabled stock on AMD boards.

 

[Makaveli]

That the OP's SSDs are, "junk," is not a fact, the OP has made no mention of BSODs, which are hardly ever an issue nowadays. The only facts were RMA rates, and that the performance would be lower than more expensive SSDs (and that's part of why those SSDs could command higher prices and fewer deep sales in the first place!), and a little slower than an Intel drive controller. Both of these are things the OP aught to have known, but both of them also should only account for 10-20%, not 50%.

Most importantly, the RAID is not improving sequential performance, which should have zero to do with the performance/quality of the individual drive, and the OP already has the hardware, and has no apparent problems with it, except for the lack of improved performance in RAID, so far as we know.

I wouldn't get one of those OCZ drives, either, but for the OP's issue, it is pointless BS, and little more than thread-crapping. He should leave that to hot deals threads on OCZ drives, and SSD recommendation threads.

I am interested to know if write-back caching being off by default was the main culprit (read-ahead can affect simple benchmarks and real workloads very differently, sometimes, so I can understand it being off by default, even if it were to lead to lower sequential read benchmarks).

I never said they were junk that was the other guys post.

The only thing I agreed with in his post is the RMA's numbers also.

I agree with anything else in this post.

 

[Cerb]

I never said they were junk that was the other guys post.

Yes, and that's what I was replying to and about; that wasn't towards you.

With fair stripe sizes, write-back caching can offer a huge performance difference over write-through. If the OP can't enable it, as Sequoia464 suspects due to lack of BIOS updates, or if that doesn't do anything, at the least it will be highly useful info. If it can be turned on and fixed things, that'd be awesome for the OP, and a record of it may help others. If not, it may be able to serve as a warning, and be used to poke and prod AMD and/or their board partners into fixing it.

 

[Grimshad]

Performance of an SSD is determined by how old it is or rather how many times it has been cycled. If the drive is brand new then there may be other factors, but most low SSD performance is because the drive has gone through its first cycle. After this point drive performance greatly diminishes. Also the larger the SSD, the slower and less reliable it is. Current SSD tech is inherently flawed and requires clustered blocks to be written in full instead of blocks at a time. This slows down the speed of SSD drive and seriously diminishes its remaining life. Best to use raid hybrids until SSD flash memory is replaced with something that works correctly like 3D memory. Flash memory will be dead within the next 5-10 years due to inability to scale.

In the end, currently SSDs are a waste of money because you will need to replace them sooner then you think.

 

[Cerb]

There's just one problem: it's not true. Even drives that have been written over to their rated cycles perform almost like new. The more complicated mapping tables causes a slight slowdown, but generally <10%, except on SF drives. Real-world WA tends to around 1.1x for non-SF drives, though I prefer to think about WA by rounding up to the next integer, since averages over too many users, often with ideal workloads, can be deceiving. For use cases where WA is a problem, drives can use slightly different WA algorithms, and rely on much more reserved space. So, really, it works out just fine (IE, those users should know who they are and not buy consumer drives with 7% OP).

Flash will run out of steam, yes, but just as you note, the major players in the market are working on suitable future derivatives. Just as they are working on keeping erase cycles from lowering too much as the flash shrinks. Worrying about SSD tech now is like worrying about HDDs not scaling if perpendicular recording were to have failed. Somebody will come up with something that works, out of the many different R&D efforts going on.

In the end, currently SSDs are a waste of money because you will need to replace them sooner then you think.

There are too many users with crappy SSDs that are alive and kicking to give that much heed. HDDs typically have a service life of 3-5 years, and outside of embedded devices, SSDs will have similar useful lifespans. By the time they get replaced, newer better technology will be available. It's not like anyone replaces the NAND on the board, or anything.

 

[Zxian]

@Grimshad - Here's a link of an in-depth test for the write endurance of a consumer level drive. As you can see, even after writing 270TB (yes, terabytes) to the drive, the performance stays pretty much on par with the fresh drive. I'm fairly confident that I haven't written 270TB of data to the sum total of all my drives, mechanical or SSD, that I've ever owned.

As Cerb mentioned, storage solutions are going through a large transitional phase. SSDs are sufficiently fast and have acceptable lifespans for most needs. Hard drives densities are starting to plateau as well, and with the adoption of cloud storage and streaming services, the amount of local storage that we'll need is likely not going to increase drastically in the near future.

 

[ParseMeHard]

LOL, a lot of fanboys in this thread. I agree, Intel>AMD any day. Only reason I'm still running this current rig, is I don't need anything more. It does exactly what I need it to do, and maxes out the few games I play on it. Reason I RAID'd these in the first place, is I got the first drive from a friend for $30 unopened. Figured I'd grab a new one for $80 and give it a shot..

Anyways,

No, I was not able to change the settings. They aren't grayed out, but I cant click them.

After letting the PC sit for about 15 hours of no use, I ran this benchmark after about 5 reboot cycles...

Now, what is wrong

I have a feeling I'm going to need to reset these drives and upgrade the RAID driver and start fresh, though, I don't see that fixing my problem.

Ill have to pick up some SATAIII 6.0 cables, though i already tried two other sets.

If I wasn't looking at buying a house, I would pick up an LSI Card and get two more of these drives. Newegg sent me a 20% off code for all RAID controllers today... I was like, WTF...


Anyways, so what has happened to the array now?

 

[sequoia464]

i had a similar Gigabyte board for a while - it had some oddities if I remember correctly. I think I had to enable raid5 as well as enabling sata III in the bios to get it to perform correctly.

If I remember correctly, Sata III isn't enabled by default in your bios.

 

[Rubycon]

In the end, currently SSDs are a waste of money because you will need to replace them sooner then you think.

Pure FUD.

@Grimshad - Here's a link of an in-depth test for the write endurance of a consumer level drive. As you can see, even after writing 270TB (yes, terabytes) to the drive, the performance stays pretty much on par with the fresh drive. I'm fairly confident that I haven't written 270TB of data to the sum total of all my drives, mechanical or SSD, that I've ever owned.

270TB is just scratching the surface. :biggrin:

The Samsung 830 in this thread: http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm has written well over 4PB and is still going strong!

Bottom line: If you NEED strong sequential performance you NEED a dedicated smart host, not the motherboard's "solution". If just running a pair of drives, it may suffice but adding more drives is going to disappoint and is for the most part is platform independent. In my experience with AMD (nforce) the experience was less than stellar. With the debut of Conroe in 2006 I've never looked back.

 

[Zxian]

@ParseMeHard - If you're willing to go through the hassle of reinstalling and testing, what would be a really telling is if you leave the controller configured in RAID, but leave the drives as two independent drives. Install Windows onto one of them and run benchmarks on each of them.

If you end up only getting ~250-300MB/s on each drive, then for some reason, your controller is kicking the drives into SATA 3Gbps. If each drive independently can reach the ~500MB/s reads that it should, then we've got something else going on.