Jump to content

System Testing and Benchmarks


NoelC

Recommended Posts

These are some Aero Glass benchmarks with 8.1 from my SSD drive using "AS SSD Benchmark" with Aero Enabled and also Aero Disabled.

I love Aero Glass but as you can see there is a significient difference anyone else have similiar results or comments?

 

~DP  :whistle:

 

ENABLED

2263_With_Aero.png

 

DISABLED

2888_No_Aero.png

 

 

 

Link to comment
Share on other sites


Thing is, AS-SSD tends to vary a LOT from run to run anyway, which I've discussed above. That's one of the reasons I don't use it to try to detect system issues.

 
As mentioned earlier, also with Intel drivers your RAM write-back cache affects the numbers quite a lot in that case, and varying amounts of available RAM for cache at the instant you're doing the testing can cause even more variation in results.

 
Here are my AS-SSD results with Aero Glass on or off (no Intel RST drivers, nor RAM cache variations), as tested just now...

 
Aero Glass for Win 8.1 On:
ASSSD_With_AGW8.png

 

Aero Glass for Win 8.1 Off
ASSSD_Without_AGW8.png

 

As you can see, my results differ a fair bit from run to run as well - but in the opposite direction from yours.

 
-Noel

Link to comment
Share on other sites

 

Only thing is I'm not sure I want a WEI score badly enough to trust it not to have a malware payload hidden somewhere inside.  I have stored copies of winsat output and can easily compare them looking for evidence of problems.

 

I don't think that it is malware, as for the results and the rest stored information: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Winsat

Link to comment
Share on other sites

I don't see detail on how much of the low-level flash management is moved to the CPU.  Dedicated controllers do the job of managing the hardware inside the SSD boxes with today's models.  If NVMe moves much of that task to the driver, not only might that take resources away from your applications but it's not hard to imagine that data could be corrupted upon any PC instability.

 

Given that the market penetration percentage of Windows 8+ is deep in the single digits, TELVM, do you think Intel might be working on Windows 7 drivers for the thing?  :yes:  Bet on it.

 

I would like to see it compared to something like RAID 0 4 x 512 GB OCZ Vector or 4 x 512 GB Samsun 840 Pro SSD arrays using traditional SATA III.  I've seen benchmarks with such hardware that compete favorably with the numbers they're posting, and without quirks (such as 344 BYTES per second with 512 byte writes - what's up with that?).

 

The price seems pretty attractive, considering the $1.50 to $3,02 / GB range is list price for a newly announced part.  I wonder what they'll drop to after they ramp up production.

 

I'll bet server farm managers are salivating over this.

 

-Noel

Link to comment
Share on other sites

I would like to see it compared to something like RAID 0 4 x 512 GB OCZ Vector or 4 x 512 GB Samsun 840 Pro SSD arrays using traditional SATA III. 

 

Over on anantech.com, they state:

 

"A single P3700 ends up replacing 4 - 6 high performance SATA drives."

 

-Noel

Link to comment
Share on other sites

Ouch ouch. From Anandtech:

 

randomreadsm.png

 

"Small block random read operations have inherent limits when it comes to parallelism. In the case of all of the drives here, QD1 performance ends up around 20 - 40MB/s. The P3700 manages 36.5MB/s (~8900 IOPS)"

 

 

^ This breaks the party as far as I'm concerned. Any good SATA III SSD can manage those 4K reading speeds @ QD1 .

 

There seems to be some sort of hard to pierce 'sonic wall' precluding huge leaps in random 4K reading.

Link to comment
Share on other sites

You're right, 4K is where it's at to improve responsiveness.

 

Re: the wall, I've always figured what must be happening is that the entire system takes a certain finite minimum amount of time to do an I/O.  At 4K bytes per chunk, you have to complete tens of thousand I/O operations per second to get better than 40 MB/sec throughgput.  That means whole operations have to complete, end-to-end, soup-to-nuts, getting 4K of data in and out literally in 100 microseconds or less (and do it over and over again consistently, not just once). 

 

As one who was around when microprocessors could execute only four or five machine instructions in that time, I always find that impressive in a relative sense.

 

The key of course to getting great small I/O performance is to minimize the latency added by the drive hardware itself, and that's what modern SSDs do extremely well. 

 

The Intel NVMe tech, on the other hand is apparently supposed to help with the latency in the pipeline.  I don't think we've seen all that it has to offer just yet, and I suspect it's going to help with that "sonic wall".  Perhaps Intel is purposely holding back, figuring as long as they can improve the speed and post bigger and bigger performance numbers, they won't need to drop the price.  Marketing.  No sense in showing their hand all at once.  Every new technology always benefits from optimization.  Just look at the differences between the drivers supplied by Windows and the latest versions downloaded from Intel.

 

By the way, a user on the OCZ forum not too long ago posted ATTO results that showed his RAID 0 array of Vector SSDs, using a modern motherboard and Intel RST subsystem (which caches low level I/O extremely effectively) delivering over 800 MB/second for 4K I/O.  ATTO uses Queue Depth 4 by default.  Edit:  I'm still trying to find that info.

 

-Noel

Edited by NoelC
Link to comment
Share on other sites

NoelC, I know this is OT, but I was wondering ...
 
Since you admit that Win8.x is slower than Win7SP1 and there is not really much, if anything, that is a real improvement for actual work in Win8.x, and you have had to add several third party apps to Win8.x to bring back the look and feel of Win7SP1 to the desktop that seems generally preferred, have you considered doing a new install of an updated Win7SP1, complete with all of your old tweaks and anything new you have learned over the last two years along with updated versions of all the apps that you normally use day-to-day, maybe even set your system up to be able to dual boot Win7 and Win8.x?  You could then truly show apples to apples performance comparisons, say which apps perform better with one OS over the other, and maybe even make a few updates to the Win7 version of your book if appropriate, since I'm sure that Win7 will continue to have life and followers for quite awhile.  I'm sure that xpclient could also add comments about the comparison as well. Just a thought since you seem to be about the most capable person I know to do such a comparison on real metal.  I mean I know your are busy with your actual job and RL, but in your spare time ... heh-heh :)

 

NOTE: You might have already made such an installation, but I don't know if you have gone back and updated your Win7 book accordingly, if anything new was found, that is.
 
[ /OT ]
 
Cheers and Regards my friend

Edited by bphlpt
Link to comment
Share on other sites

I'll come right out and say that there is literally NOTHING I've found essential in Win 8.1 that I couldn't have or do in Win 7.  Only thing is, it's current - and that may matter for things like security fixes, driver updates, etc.

 

I still maintain a VM that's the best setup I had done with Win 7, as configured per the best practices I listed in my book, and I have kept the system up to date.  I ran it just a little while ago to ensure that even with the latest updates the file system is a fair bit faster than Win 8.1.  I also maintain a Win 8.1 VM that's just like my host system, with an identical virtual hardware setup and configured to the best of my abilities, so I've been able to do some pretty careful head to head comparisons.  I don't think they're any less valid or useful than two different systems running on the same hardware, but with necessarily different hardware drivers.

 

In short, my well set up Win 7 VM is a good bit faster than my well set up Win 8.1 VM.

 

I haven't had to tweak my Win 7 book much in the past few years because it is essentially complete - it takes one from an out-of-box experience to a fully functional, well-tuned workhorse of a system.  Since I don't use Win 7 any longer for everyday work, I probably don't pay as much attention to the subtleties of Win 7, and I've put more effort into making my Win 8 book better, though to be fair I HAVE added a few sections to the older book in the not too distant past.  Looking at the market now - with Win 8 basically flopping badly - I probably SHOULD put some more effort into the older volume.  Thank you for the suggestions.

 

-Noel

Link to comment
Share on other sites

... By the way, a user on the OCZ forum not too long ago posted ATTO results that showed his RAID 0 array of Vector SSDs, using a modern motherboard and Intel RST subsystem (which caches low level I/O extremely effectively) delivering over 800 MB/second for 4K I/O.  ATTO uses Queue Depth 4 by default.  Edit:  I'm still trying to find that info.

 

I'd expect that setup to hit the same 'sonic wall' @ 4K random read QD1.

 

4x Samsung 840 Pro (among the fastest consumer SATA III SSDs) in raid 0:

 

2a5f3152_as-ssd-benchIntelRaid0Vol6.2.20

 

^ Everything scales up nicely except 4K random read QD1, which stalls at almost exactly the same point that the Intel P3700 beast above.

 

 

 

"4K random QD1:

 

At the low queue depths you normally encounter in a desktop environment, all of these configurations perform fairly similarly. In fact, the striped setups are even a bit slower than single drives. This is because we're taxing the NAND flash's throughput. Parallelism is needed to distribute the workload across multiple dies on multiple channels.

 

4K random QD64:

 

Once we jump up into very high queue depths, both RAID-based arrangements distinguish themselves. It's only a shame that this is very atypical of any desktop workload, so you won't see it unless you take the 840 Pros into a more enterprise application.

 

... ...

 

... Striped drives are certainly better equipped to push more IOPS, but only when you're stacking commands more than four high. Jumping up to a queue depth of 32, 16, or even eight is really uncommon in a desktop or workstation environment. As a result, the performance differences are far less pronounced in the real world.

 

One SSD on its own scores again in the contrived tests we put together. The performance differences when we boot up and shut down Windows 8, then fire up different applications, are marginal at best and not noticeable in practice. Single drives actually manage to outperform the striped arrays some of the time, even ..."

 

 

Uncle Tom's - SSDs Raid 0: Great for becnhmarks, not so much in the real world

Link to comment
Share on other sites

 

I'd expect that setup to hit the same 'sonic wall' @ 4K random read QD1.

 

That's the speed of modern computers for you.  Hey, I envy 35.98 MB/sec since I get about 26.

 

Note in the AS-SSD output above the small latency times reported - mere tens of microseconds.  And of course the write times are representative of no-wait RAM write-back cache, which tends to batch the writes together to the actual drive.  And speaking of cache, don't discount the effect it has on both writing and reading.  Benchmarks don't actually do that good a job showing you how well it works in real-world operations.  The bottom line is:  If you have a fast I/O system and lots of RAM, you will have a very responsive system for real world operations even if your 4K QD1 random I/O benchmarks aren't gargantuan.

 

Interesting read, that Tom's article.  Thanks for the link.

 

People sometimes don't realize that SSDs generally have so much higher MTBFs than HDDs that a RAID array of SSDs can be as reliable as a single electro-mechanical HDD.

 

And everyone should be doing backups.

 

After 2 years of practical experience I'm glad I went in the direction I did (4 SSD RAID 0) and would tend to do it again today (except I'd start with modern drives with low latencies).  Next system I put together I'll certainly consider those Intel boards, though.  It will be interesting to see what Samsung comes out with in that arena in the near future.

 

-Noel

Link to comment
Share on other sites

People sometimes don't realize that SSDs generally have so much higher MTBFs than HDDs that a RAID array of SSDs can be as reliable as a single electro-mechanical HDD.

Care to expand/detail on this? :unsure:

Like some actual MTBF comparison, detail on the Raid level used, etc?

And everyone should be doing backups.

Sure :yes:, the three golden Rules (JFYI):

http://www.msfn.org/board/topic/128807-the-solution-for-seagate-720011-hdds/?p=930329

jaclaz

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...