Licensing question re: 2012 R2 Essentials and IIS
-
@MattSpeller said:
@creayt at least account for the rest of it lol - add the cost of the RAID controller to the 10 and RAM to the single
What's going to be epic is when Samsung releases firmware and software updates and adds support for Rapid Mode across a software RAID of their SSDs and supports infinite cores, that'll be a game changer. It maxes out at one drive and something like 8GB of used RAM at this point I think.
-
@creayt that would be very interesting! Have they announced any plans etc for that?
-
@creayt said:
@mlnews said:
@creayt said:
I don't think you can make a single SSD push 4.5 giggers a sec on a *Nix OS, so my raw performance would be much worse.
Where is Windows getting that performance from that UNIX doesn't have?
Samsung's Rapid Mode on any 840/850 SSD. It turns the system's RAM into a write back cache and this box has 32GB and gets this performance:
That mode is only needed on Windows because Windows doesn't do that natively. UNIX does that without special software. If something has "better performance on Windows" that should be a red flag that something is being missed. UNIX is used for the highest performance, most demanding environments. Outside of video gaming, it should be really shocking to find any UNIX system that doesn't keep up or crush Windows in performance.
-
@scottalanmiller said:
That mode is only needed on Windows because Windows doesn't do that natively. UNIX does that without special software. If something has "better performance on Windows" that should be a red flag that something is being missed. UNIX is used for the highest performance, most demanding environments. Outside of video gaming, it should be really shocking to find any UNIX system that doesn't keep up or crush Windows in performance.
We're talking about Samsung's Rapid Mode software layer, it's not part of Windows. It's written for and only supported on Windows, because that's their market. If you're saying there's a Unix-available equivalent, what's it called and what kind of numbers can it get out of a single SSD?
As far as I know there's no way to get anywhere near that ballpark of performance even on Unix, OS X, or Linux at the moment at least.
-
@scottalanmiller said:
UNIX does that without special software.
Wait, so are you saying that Unix, by itself, uses all available system RAM as a write back cache for all applications blindly?
-
@creayt said:
We're talking about Samsung's Rapid Mode software layer, it's not part of Windows. It's written for and only supported on Windows, because that's their market. If you're saying there's a Unix-available equivalent, what's it called and what kind of numbers can it get out of a single SSD?
I know, that's what I was explaining. Samsung is making third party code to bring into Windows something that every major competitor has natively. It's not called anything, it's just how UNIX works
It's just the ram cache.And it can get whatever you can get out of memory performance. It's a RAM cache.
-
-
@scottalanmiller said:
@creayt said:
We're talking about Samsung's Rapid Mode software layer, it's not part of Windows. It's written for and only supported on Windows, because that's their market. If you're saying there's a Unix-available equivalent, what's it called and what kind of numbers can it get out of a single SSD?
I know, that's what I was explaining. Samsung is making third party code to bring into Windows something that every major competitor has natively. It's not called anything, it's just how UNIX works
It's just the ram cache.And it can get whatever you can get out of memory performance. It's a RAM cache.
Wow, thanks. I'm about to Google my ass off. Does that mean that Unix in general is more likely to lose data in the event of power loss than Windows?
-
@creayt said:
Wow, thanks. I'm about to Google my ass off. Does that mean that Unix in general is more likely to lose data in the event of power loss?
Yes, because UNIX is mostly designed for enterprise class gear where power loss is something you are supposed to protect from the outside. UNIX has enterprise software RAID too, same issues. But it is configurable, so not a real issue.
Oracle makes a big point of this.... instead of building power protection inside the chassis like hardware RAID does, they expect you to put that outside the chassis.
-
Have not read this yet but likely this has info that you want...
http://www.zfsbuild.com/2012/04/18/let-zfs-use-all-of-your-ram/
-
@scottalanmiller said:
Have not read this yet but likely this has info that you want...
http://www.zfsbuild.com/2012/04/18/let-zfs-use-all-of-your-ram/
Read that, seems like it was just pointing out a lack of foresight in the initial design of ZFS and how it arbitrarily ignores a heavy chunk of RAM resources at any scale. Are you saying to look into ZFS itself?
-
I wonder if part of the reason the Samsung version benchmarks so high is because of these subtechniques ( which for all I know may be copied from or already exist in Unix
Read/Write Cache. RAPID mode uses system DRAM as a cache of “hot data” based on frequency, recency, file type, etc, such that subsequent requests can be served directly from DRAM, rather than going to the SSD.
Write Optimization. System write requests are processed for optimized performance with [ the Samsung hardware itself ].
File Awareness. RAPID mode may exclude certain files from caching based on a variety of factors, including file type, file size, etc. This prevents unnecessary data from polluting the cache.
Persistent Cache. RAPID mode maintains cache map across system reboot to maintain consistent high performance
operation.Cache Compression. RAPID mode dynamically compresses and de-compresses cache contents to dramatically improve cache efficiency. Optimized for Samsung hardware. RAPID Mode was co-developed and optimized for the
Samsung MEX controller.Does anyone here have a Unix box available w/ a single SSD they can run something Crystal like and see what numbers it puts up? I'd be curious to see how close to my numbers on Windows on my ~$1,000 system it can get using the Unix built-in stuff.
-
@creayt said:
@scottalanmiller said:
Have not read this yet but likely this has info that you want...
http://www.zfsbuild.com/2012/04/18/let-zfs-use-all-of-your-ram/
Read that, seems like it was just pointing out a lack of foresight in the initial design of ZFS and how it arbitrarily ignores a heavy chunk of RAM resources at any scale. Are you saying to look into ZFS itself?
No, it was a fast search. I'm in the middle of a big database migration project and don't have much time to respond. ZFS is just one filesystem on UNIX that has some extensive RAM caching functionality.
-
And as you can see because of the silly way that Mac OSX identifies running windows, I can't keep track of my own identity today.
-
@mlnews said:
@creayt said:
@scottalanmiller said:
Have not read this yet but likely this has info that you want...
http://www.zfsbuild.com/2012/04/18/let-zfs-use-all-of-your-ram/
Read that, seems like it was just pointing out a lack of foresight in the initial design of ZFS and how it arbitrarily ignores a heavy chunk of RAM resources at any scale. Are you saying to look into ZFS itself?
No, it was a fast search. I'm in the middle of a big database migration project and don't have much time to respond. ZFS is just one filesystem on UNIX that has some extensive RAM caching functionality.
NP, thanks for the help. Sounds like a/the major competitive advantage for Samsung Rapid Mode is that they write the software and make the hardware and can directly optimize the software for the hardware ( deducing this as it was alluded to many times, and it makes a lot of sense ).
-
@scottalanmiller said:
And as you can see because of the silly way that Mac OSX identifies running windows, I can't keep track of my own identity today.
But I admire that you can get work done on OS X at all. To me it feels like dropping back into a Pentium 4.
-
Looks like it does heavy analysis of user-access patterns and does speculative loading. Thus, it may inadvertently ( or purposefully ) game benchmarks.
-
@creayt said:
Looks like it does heavy analysis of user-access patterns and does speculative loading. Thus, it may inadvertently ( or purposefully ) game benchmarks.
Benchmarks are a game as it is. By that I mean that how a storage subsystem performs for any given workload is unique. So a benchmark, no matter what it is, can only represent one workload type. And so any given subsystem can only be truly tuned for a single workload type. So it is a game system no matter what, in a way.
-
Tiering is handled much the same way. If you get Drobo B1200i with tiering, it looks for usage patterns and moves the most used blocks onto the SSD tier. Same as a RAM cache. This is generally want you want - a system that games itself to look good to you.
-
That ZFS article is a bit weird. 8GB was not a monster server when ZFS was new. They are looking at the memory sizes of little 32bit IA32 installs, which was not even an option when ZFS was invented. It was invented on Sparc64 only, the port to AMD64 was later.
32GB was common by the time that ZFS was made and very quickly they were rolling it out on Thumper which was AMD64 and went to 64GB. Thumper was the first place that ZFS was really used. So any system getting ZFS was either a Sparc64 system where 64GB was impressive but not shocking or an AMD64 system that was designed for 64GB out of the gate.
