New System Build - Reality Check

Ron AKA wrote:

I read that paragraph. No change in my opinion. 1 SSD will provide some marginal performance gain. 2 will provide nothing extra over 1 SSD, other than another piece of hardware getting old.
I support the second statement, but the 1st is just wrong. The OP is spending over 2 grand on this system and to cut the SSD to save $200 is about the worst case of pennywise, pound foolish there is. No one who replaced their primary drive with an SSD has gone back to spindles.
 
My main computer is liquid cooled and I don't hear it run while sitting next to it. The only way I know computer is on is the red light on the case.

The key is low speed but large and high volume fans. Also, keep fans away from blowing on to surfaces.
No insulation whatsoever.

A with Pugent I did not specify insulation or special parts and I doubt very much if they include it in the system unless specified.
 
This is got to be something very new because I have never heard about it before.
Wiki indicates that two separate controllers must be used. I am wondering if that guy who setup Raid 1 with higher read performance used two controllers and then used software RAID 1 to mirror the drives.

I don't think using one controller is possible for RAID 1 and better read performance.
 
Relax, you will have satisfaction of building the system yourself. :-)
 
SushiEater wrote:

My main computer is liquid cooled and I don't hear it run while sitting next to it. The only way I know computer is on is the red light on the case.

The key is low speed but large and high volume fans. Also, keep fans away from blowing on to surfaces.
No insulation whatsoever.

A with Pugent I did not specify insulation or special parts and I doubt very much if they include it in the system unless specified.
It's a clear part of the Serenity line, which can get down to 11 decibels.

"We use thermal imaging cameras to view the real time heat dispersion inside our computers and then tune the location and speed of our fans to give the best silence vs. performance levels. We custom program the fan RPM curves for optimal temperature response. We cherry pick the most quiet of components from our inventory. Then we back it up with our industry leading warranty. In the world of quiet computers, there is simply nothing (including completely passive systems) that can top the performance and silence of the Serenity."

That's the sort of thing one might be willing to pay money for. Maybe even a lot. Or again, settle for the 90 or 95% for considerably less. It's a free world, unlike with the Apple side where you only have unauthorized approaches.
 
SushiEater wrote:

This is got to be something very new because I have never heard about it before.
Wiki indicates that two separate controllers must be used. I am wondering if that guy who setup Raid 1 with higher read performance used two controllers and then used software RAID 1 to mirror the drives.

I don't think using one controller is possible for RAID 1 and better read performance.
Sure it is. The bottleneck is going to be the physical drives, not the I/O via the controller and SATA ports.

With a modern Motherboard with two drives in a mirrored config on separate SATA ports, the CPU is going to be the biggest bottleneck as far as performance increases via RAID 1.

But, with a reasonably fast CPU, just using software mirroring with Windows should give you a pretty nice performance increase in read speed (as in the benchmarks someone performed using that technique that I linked to). Again, because the exact same data is on both drives in a mirrored config, the OS can service multiple requests at the same time (one request from one drive, at the same time it's servicing a different request from the other drive).

This is not a new thing. I wrote a number of benchmarks (DOS programs) many years ago to get a better idea of how different drive configurations worked with Novell Netware (older 2.x and 3.x releases of it) while working for Sprint Long Distance.

My findings were that setting up drives in a mirrored configuration (even using the same SCSI Controller card versus separate cards), resulted in a huge increase in performance. The bottleneck was the drives themselves, not the controller cards.

Interestingly, even in configurations where you had a lot of write activity, performance still increased significantly with a mirrored drive configuration (because even with apps that write more than read, you're still going to have tons of read activity for getting to the correct location you're writing to, getting to where you need to apply locks, etc.).

Using disk mirroring resulted in significantly better performance in virtually any configuration and application combination, even using apps that had a lot of write activity.

With Windows, the results are not as consistent, and Windows has never had very good disk algorithms for managing read/write activity compared to much older Operating Systems like Novell Netware to begin with (Novell Netware disk algorithms were always dramatically better -- even using much older versions of it). Heck, I've seen Novell Netware 3.x running on a slower 386 CPU "run circles" around NT running on much faster Pentium models with even more memory.

The results were not even close (Netware running on a very slow CPU would be many times faster than NT running on a much faster CPU).

Microsoft's disk algorithms have always been horrible (and using the term horrible is being nice to them) compared to some of the other Operating Systems.

But, over the years, Windows disk algorithms appear to be improving from some of the benchmarks I've seen lately. So, I'd expect to see a pretty good performance increase overall with mirrored drives with Win 7 versus a non mirrored config (at least using Windows software for mirroring with a fast CPU, although I'd test semi-hardware based configs to see what you get).

--
JimC
------
 
Last edited:
And where did you find that? RAID 1 is nothing more than a mirror.
Exactly. And therefore reads are twice as fast.
You need RAID 5 (and one extra HD to improve read performance.
There is no need for any extra HD's beyond the RAID 5 array to improve performance. RAID 5 performs very well as is.
Nope, Raid 5 read performance is better because of the striping but the writes will be slow.
The striping AND the number of drives is why RAID 5 performs so well. Reading from four or more drives in parallel is much faster than reading from one. Writes will be somewhat slower due to parity calculations and writes, but I'd hesitate to call them slow. Especially SSD RAID 5 systems are blazingly fast, saturating workstation IO.
For both read and write performance and redundancy you need RAID 10.
That's really just RAID 1 over multiple drives.

Jesper
 
kelpdiver wrote:
Ron AKA wrote:

I read that paragraph. No change in my opinion. 1 SSD will provide some marginal performance gain. 2 will provide nothing extra over 1 SSD, other than another piece of hardware getting old.
I support the second statement, but the 1st is just wrong.
I disagree with both. An SSD is a must for a modern system. It makes such a vast difference it's almost unimaginable.

But a RAID 1 pair of SSD's is in another league entirely. Truly amazing performance.

Jesper
 
kelpdiver wrote:
No one who replaced their primary drive with an SSD has gone back to spindles.

kelpdiver wrote:
No one who replaced their primary drive with an SSD has gone back to spindles.
Really? I auditioned Win8 on both a Samsung 830 256GB and a 250GB Velociraptor (same computer so other hardware differences were zero). Aside from initial boot and initial app loads, the SSD has no significant advantage over the Raptor in my environment.

Since I reboot once in a blue moon, boot speed is a moot point. Once Win8 has cached everything my app load times are surprisingly quick with the Raptor, in the 2 to 5 second range for the apps I use — including LR and Photoshop. System tools (File Explorer, IE, etc.) load virtually instantly.

Bear in mind too that once your apps are loaded in RAM, they operate from there, not from your disk. Since I will spend hours in Photoshop, what do I care if it loads in 5 seconds or 1? For apps that do require frequent disk IO, an SSD can be employed to serve that purpose, at least in my case as I explain below.

So given that the spinner was quite capable vis-a-vis the SSD, I left my OS and apps on it and gave the SSD the LR catalog and ACR cache, PS scratch, my working images folder, Color Efex Pro 4 virtual memory, NX2 thumbnail cache and various and sundry other sorts of temp files that would benefit from the SSD's qualities.

I know that this goes against conventional wisdom and my results are probably as good as they are due to a large amount of RAM the the fact that I tend to run the same dozen apps or so pretty much exclusively. I specifically am NOT advocating that anyone who has an SSD as their system disk should revert back to the older technology. What I am saying is that if you have a fast HDD and plenty or RAM, the SSD advantage is not as great as some would have you believe.

I need to make a video... :-)

Adobe's thoughts on PS and SSD
 
Last edited:
Ho72 wrote:

I know that this goes against conventional wisdom and my results are probably as good as they are due to a large amount of RAM the the fact that I tend to run the same dozen apps or so pretty much exclusively. I specifically am NOT advocating that anyone who has an SSD as their system disk should revert back to the older technology. What I am saying is that if you have a fast HDD and plenty or RAM, the SSD advantage is not as great as some would have you believe.
sure - throw a lot of memory at the problem and never reboot, and it really doesn't matter how good the random access performance of your disks are. The Raptors are only a tiny improvement in this regard on 7200 or even 5400 rpm drives. It's the memory doing the work here.

but it's most typical for people to turn off their power systems when not in use.
 
Doublehelix wrote:

Dang Jim! I wish I would have read that message earlier. I have already bought all the components and have already started the build... (mobo, memory, PSU, CPU, DVD burner already installed). Should finish the rest tonight.
You can actually find prebuilt systems using most of the components you want for even less from some sites.

But, AVA Direct has a better reputation for their builds (burning in systems to make sure all components work together, etc.).

So, for a premium system build like you're putting together now, AVA Direct can be a good choice (so you'd have tested components, warranty, etc.; versus having to worry about RMAs for parts that have problems and issues putting together the same components yourself)

They've got a very good reputation for offering premium systems at a good price. After all, they're buying parts at dealer prices, so they can get them for less than you can and still make a profit on systems, even after assembling them and offering a 3 year warranty.

My apologies. I should have mentioned that option earlier in the thread (but, I didn't' price out your system with them until you had already settled on a "final build", realizing at that point that you could probably get the same system for about the same price already assembled and including a 3 year warranty on parts and labor).

--
JimC
------
 
Last edited:
I don't think mirrored drives have any chance of doubling transfer rate. Reads will speed up some, and writes will slow down some.

In the old days of Novell servers, the software was dedicated to that purpose using mirrored drives. As a server it could be writing to one drive and reading from the other, in a multi-user environment, and it really was faster. I doubt on a home PC that a mirrored drive is that smart, and the workload is different to a multi-user server.

On a home PC and sticking to reasonable level of complexity, I would suggest a Raid0 could be the most cost effective. First you double the disk capacity as Raid0 effectively joins the disks. That reduces cost. Of course there is no redundancy built in. The obvious solution is to use and external network drive that does your backup in the background. The in computer disks do the heavy lifting for the application that needs the file access.

Here is an article which found that Raid0 actually does double disk transfer rates. However that does not mean it doubles application speed. Those benefits were more in the range of 5-10% reduction.
 
Ron AKA wrote:

I don't think mirrored drives have any chance of doubling transfer rate. Reads will speed up some, and writes will slow down some.
Benchmarks for windows have been provided. I might have two spare 250gb drives to use to generate some zfs benchmarks for the 3 situations.
Here is an article which found that Raid0 actually does double disk transfer rates. However that does not mean it doubles application speed. Those benefits were more in the range of 5-10% reduction.
Striping definitely increases large file sequential reads/writes. If you're copying ISOs back and forth between two sets of R0 disks, it's a no brainer. But if you're trying to go to the network on a 1gbit interface, the extra does you no good.

The gotcha is that for random access, striping does not benefit, and that's where most of the action is. And that's why that 100% drops to 5-10, despite doubling the risk of total failure.
 
kelpdiver wrote:
But if you're trying to go to the network on a 1gbit interface, the extra does you no good.

The gotcha is that for random access, striping does not benefit, and that's where most of the action is. And that's why that 100% drops to 5-10, despite doubling the risk of total failure.
Backup does not have to be done in real time, and does not have to impact real time application disk performance. In 20 years of running pc's at home I've had one mother board failure, one power supply failure, and no disk failures.
 
Ron AKA wrote:

I don't think mirrored drives have any chance of doubling transfer rate. Reads will speed up some, and writes will slow down some.
RAID 1 on SSD consistently doubles read rates, except on very small file operations during low IO load. On HDD it's a bit more complex but read rates are higher during IO load, when it matters most.
In the old days of Novell servers, the software was dedicated to that purpose using mirrored drives. As a server it could be writing to one drive and reading from the other, in a multi-user environment, and it really was faster.
That ... is not how it works.
I doubt on a home PC that a mirrored drive is that smart, and the workload is different to a multi-user server.
Why is it "not smart" to have data redundancy and higher speed, and minimal complexity increase? You merely assert this and provide no reasoning.
On a home PC and sticking to reasonable level of complexity, I would suggest a Raid0 could be the most cost effective.
With the failure rate of consumer hard disks, I would suggest that solution is insane!

Jesper
 
Backup does not have to be done in real time, and does not have to impact real time application disk performance. In 20 years of running pc's at home I've had one mother board failure, one power supply failure, and no disk failures.
In 30 years of running PC's at home I have had about half a dozen mother board failures, around as many PSU failures and dozens of disk failures, all of them sudden, and all of them with at least some data loss.

I have lost count of how many acquaintances have come by with failed disks which I have tried to rescue data from. Sometimes I can get something, but often I have to disappoint them.

There is no way I will trust my data to a spindle drive even short term. Backups of anything important have to be done now.

Jesper
 
kelpdiver wrote:
Ho72 wrote:

I know that this goes against conventional wisdom and my results are probably as good as they are due to a large amount of RAM the the fact that I tend to run the same dozen apps or so pretty much exclusively. I specifically am NOT advocating that anyone who has an SSD as their system disk should revert back to the older technology. What I am saying is that if you have a fast HDD and plenty or RAM, the SSD advantage is not as great as some would have you believe.
sure - throw a lot of memory at the problem and never reboot, and it really doesn't matter how good the random access performance of your disks are. The Raptors are only a tiny improvement in this regard on 7200 or even 5400 rpm drives. It's the memory doing the work here.

but it's most typical for people to turn off their power systems when not in use.
Client workloads aren't normally affected that much by random access times, usually being more sequential in nature (where the VR is very good). But agreed, without my copious RAM things wouldn't be the same. And, FWIW, my systems sleep when I'm not using them.
 
Jim Cockfield wrote:

My apologies. I should have mentioned that option earlier in the thread (but, I didn't' price out your system with them until you had already settled on a "final build", realizing at that point that you could probably get the same system for about the same price already assembled and including a 3 year warranty on parts and labor).
Haha! No worries! As was mentioned, I get the satisfaction of building it myself! This way, I do become intimately involved with each component, and will understand all of the ins and outs of the system.

But next time...!!! ;)
 
And just as an FYI, I will be using the Intel X79 chipset as the RAID controller rather than the Marvel chipset. I heard there was a difference.
 
kelpdiver wrote:
But also consider flipping these around - use the 128 for the OS + programs that don't let you pick the install patch, use the somewhat faster 240 for the more read/write intensive activity. The performance delta is nearly zero for reads, which is what the OS disk is 95% after install.
Dang! I can see your point here, and I have to agree that your logic is sound. There are so many options and permutations, and just when I think I have it all figured out, someone posts another thing to consider!!! LOL!

The good news (?) is that my mom fell and broke her hip (she is fine), and I have been in the hospital with her for the last day-and-a-half so the build is stalled in a half-finished state. This means I still have the option of making changes.

The box is sitting there still with the mobo, the memory, the PSU and the DVD drive installed. Nothing else since yesterday morning... This has given me time to read through the Asus P9X79 PRO manual however while I have been sitting at the hospital.

  • Win 8 Pro (Yes! Win 8!)
Your funeral! ( just kidding )
I almost returned it yesterday in favor of a Win 7 install disk... still on the fence here...

I'd miss the catalog more than the program data, esp if it's LR. But more to the point, Tom's people are totally wrong here, particularly for the 840 Pro. People have been worried about this from the beginning, and even though each new and higher density SSD should make it worse, no one is running into the problem (outside of enterprise work - usually databases). You will retire this system long, or install a larger SSD, long before the SSD hits write exhaustion. However, if we presumed that the threat were real, you're again better off using the 240 SSD for the write heavy activity as it has twice the write endurance (as measured in gigabytes written, not p/e cycles).
You are right, the LR catalog loss would be catastrophic, even with periodic backups to a platter and then to my normal backup scheme. I need to change the settings in LR to force backups on each program close.

I think you have convinced me to switch the 128 GB with the 256 GB drive...

Also note - the software that Samsung provides with the 840 pro will allow you to increase the percentage of blocks reserved. This lowers the total space available, but will improve performance as well as any longeavity concerns you may have. So consider upping it another 5 or 10%.
Not sure I get this (yet)... this is my first SSD drive. Blocks reserved for what? Are these blocks just reserved for non-use? I am not sure I am wrapping my mind around how that helps performance and longevity... I will try to do some reading up on this today when I am at the hospital with my mom...

THANK YOU SO MUCH for the great advice! Cheers!
 

Keyboard shortcuts

Back
Top