Overclock.net › Forums › Components › Hard Drives & Storage › SSD › [Official] Samsung SSD750/830/840/850 Owners Club
New Posts  All Forums:Forum Nav:

[Official] Samsung SSD750/830/840/850 Owners Club - Page 182

post #1811 of 5733
Quote:
Originally Posted by erickmiller View Post

I agree.

btw, I'm testing the Samsung 840 Pro in a SAN, an Infortrend DS B24F-R2840-4. It has an option to enable "Drive Delayed Write" in the drive-side parameters, but regardless of whether it is enabled or not, it appears the drive is "not" enabling its cache.

So, I'm asking Infortrend about this since I saw a note that indicated that their snapshot-capable SANs have a firmware that "always" disable the drive's write cache. Not good. But maybe they have a firmware that doesn't? I hope so, but I'm not holding my breath.

Right now, the performance is similar to the SSD in the Windows machine while the "Enable write caching on the device" option is "unchecked".

Eric
You need to give us AS-SSD benchmark.You drive can be in IDE mode,misaligned and so on.
http://www.filehippo.com/download_asssdbenchmark/
post #1812 of 5733
Quote:
Originally Posted by erickmiller View Post

I have a Samsung 830 in my old IBM Thinkpad T61p (running Windows 7 Enterprise) and you can see the same concern regarding write performance with and without the operating system caching enabled.

With Windows write caching enabled:


With Windows write caching disabled:


Pretty poor write performance. This SSD only has the default 7% free space, whereas the 840 Pro at the office has only a 100GB partition on the 512GB model, so lots of free space.

Eric

Whoa, hold on a moment, is anyone really looking at these results? Sequential read of 96MB/s, sequential write of 72MB/s, for a Samsung 830? That was with caching enabled?

Sure the system this was run on might have a SATA II interface, but the results are sub-SATA I. Looks like IDE mode, but even that does not explain the wretched results. What mother board does this system have? Is that an Nvidia SATA chipset on it? You can't come to any definitive conclusions regarding write caching on a system like that.

Also, ALL SSDs will suffer with write caching disabled? You have forgotten about SandForce SSDs, no cache in the usual sense.

Here's a SanDisk EX 240GB, on SATA II, write caching disabled:



Not to shabby for a SandForce SSD on AS SSD, all CPU power saving option enabled, no Prime95 thread running.
post #1813 of 5733
Quote:
Originally Posted by parsec View Post

Sure the system this was run on might have a SATA II interface, but the results are sub-SATA I. Looks like IDE mode, but even that does not explain the wretched results. What mother board does this system have? Is that an Nvidia SATA chipset on it? You can't come to any definitive conclusions regarding write caching on a system like that.

Also, ALL SSDs will suffer with write caching disabled? You have forgotten about SandForce SSDs, no cache in the usual sense.

It definitely only has SATA I enabled. There is a hacked BIOS for this Thinkpad (remember it is a notebook) that enables SATA II, but I haven't bothered.

Also note that this was just an example of cached versus non-cached where the performance is going to be sub-par not only because it's a notebook (I have no idea what chipset it has - I'll check), and has SATA I connectivity, but it is also encrypted using TrueCrypt. I only wanted to indicate that there was a "significant" difference between cached and non-cached.

I'll also check the mode in the BIOS to see if it is set to IDE. Not sure where to look in Windows, though, if there is a change that needs to be made. I thought there was something that needed to be re-installed to change this.

Good to see the numbers, though, with write caching disabled on the SandForce SSD.

The numbers that I'm most interested in are the poor results coming from using the Samsung 840 Pro in an Infortrend SAN. So far, it appears that the setting for "Drive Delayed Write" doesn't work, so the Samsung is stuck in non-cached mode no matter what is done at the SAN configuration. I wouldn't have dreamed, though, like you said, that the performance would be SO dreadful.

I'll post some more screenshots showing the results from the Infortrend.

Eric
post #1814 of 5733
I just rebooted the Thinkpad and checked the BIOS and it's SATA setting is set to "AHCP", and not "Compatibility Mode" (which I'm assuming is IDE mode).

This was a freshly installed Windows 7 instance less than a year ago, so I suspect Windows is using the correct driver for AHCP.

Just re-ran the Anvil's Storage Benchmark after the reboot and got similar numbers (cached), so it's pretty consistent.

Eric
post #1815 of 5733
Thanks. I'll take a closer look. That was the problem I encountered was the UI, I couldn't read it.
post #1816 of 5733
Quote:
Originally Posted by parsec View Post

When using Magician 4.0 with Windows 8, you can secure erase your Sammy's with the bootable USB SE option. But be sure to read the fine print in the Help section, found at the bottom of the Magician 4.0 screen. The UI in the bootable USB is a mess until you load a font, or use the command line interface provided. Instructions in the SE section of the Help section.

Thanks. I'll take a closer look. That was the problem I encountered was the UI, I couldn't read it.
post #1817 of 5733
Are these good results for my 840 pro 256GB
I ran AS SSD while running 1 thread of Prime95
Wonder if it would get better results with the Intel ahci driver 11.7?
post #1818 of 5733
As promised, the results from the Infortrend SAN, which appears to disable caching in the Samsung 840 Pro SSDs.

Note that this is a RAID 0 configuration with 8 x Samsung 840 Pro SSDs. The SSDs are connected to 3Gbps Interposer cards, allowing the SATA drives to be connected to SAS connections.

The Infortrend has 4GB of BBU cache (although only 15% is allocated to the channel that is being used, so only 0.6GB is used for read and write caching).

The test is performed in a 4 vCPU 4GB virtual machine running on vSphere 5 on an HP Proliant DL385 G6 with 64GB of RAM, 2 x 2.6GHz 6-core Opterons, and 2 x 4Gbps fiber channel ports connected to 2 fiber channel fabrics. The Infortrend is connected to a single fabric with only a single 4Gbps port connected.

As you can see, the write performance is pretty dreadful. Especially for EIGHT Samsung 840 Pro SSDs in RAID 0. With or without internal caching, it's pretty dreadful, and matched up with my tests where I disabled the internal device caching in Windows with a single Samsung 840 Pro.

I understand there are a lot of layers we're going through, but we're almost seeing better throughput on 15k drives configurations (not necessarily on random read I/O, but most definitely on random write I/O).





Eric
post #1819 of 5733
I have since confirmed that Infortrend disables the write cache on SATA-connected devices, regardless of the Drive Delayed Write setting. Why, I have NO idea... I have asked them why, but we'll see if I get an intelligent response.

Does the Samsung 840 Pro default the write cache to disabled or enabled if the operating system doesn't send a disable/enable command?

Eric
post #1820 of 5733
I dont get,it if you have 8 x 840 on that lsi card why it says 39gb only.Leave that story about write-back cache enabled because it cant hurt speeds that MUCH.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: SSD
Overclock.net › Forums › Components › Hard Drives & Storage › SSD › [Official] Samsung SSD750/830/840/850 Owners Club