9262 performance results with 3 independant drives that likely outperform RAID

IMO, sometimes RAID configurations are not a better choice over independant drives, but it depends on how you use your laptop most commonly. While I have never seen any data showing that RAID will hurt performance, here is an example configuration where 3 independant drives will be top performers.

Check out the graph below, it's and Oracle ETL data warehouse load program written in PLSQL, executing massively parallel on my old Compaq laptop, nx9600 with one 7200rpm hitachi drive




The drive is being thrashed by the parallel executions, and still cannot provide data fast enough to keep the cpu utilization above 25%. The processor could do a lot more work, if it could get the data from disk into memory faster. This is why I replaced this laptop with a new Sager.


The graph below is my 9262, Q9650 cpu, with 3 western digital Scorpio Black label drives, configured as three independant drives. Drive 0 holds the O/S, Drive 1 the Oracle application files and Oracle ETL application recovery area data files, and Drive 2 contains the Oracle ETL application database files (and for even better performance, the data files reside here on Drive 2 but the indexes for those same data files are on Drive 0 with the O/S).

=========
NOTE: Sorry, I destroyed the original graphic from the Sager test. Below is from a graph from a non-sager box, but it looks like the Sager results, so my comments can be understood, until I get a chance to run a new graph on the Sager...



=========

The drives are being hit with the same massively parallel executions, but they are working very little, and still providing data to the cpu fast enough to keep the processor utilization above 80%, exactly what we like to see in database applications running parallel execution models. This is why I bought the Sager and configured drives independantly.

My only point is that when you know what files you will be accessing and how, for your primary use of your laptop, you can probably configure 3 independant drives to outperform a RAID config. All this of course, is IMO

 

can you show us some real time apps.

like how long photoshop will take to open up, how long a games will load, how it effects start up, defraging time with first install, etc,

it might change your mind (or mines)

 

I don't spend 8 hrs a day opening photoshop, or loading a game, or rebooting, or defragging, and neither do you.

I spend 8+ hours a day compiling and executing parallel database applications, and that's what this box is configured for. Your mileage will of course vary...

 

Now that is presumptuous, perhaps he is independently wealthy and does
play games 8 hours a day...who knows. Point is, compare an apple to an apple then discuss taste, texture and seed count. Like he said, you may change your mind....or his.

Just my opinion...

 

Er, why would RAID'd drives be used to increase performance? This post essentially says "not having to write redundant drives gives better performance"! (Sort of obvious no?) The point of having multiple hard drives in a RAID configuration is reliability, not performance (RAID 0 not really being RAID).

 

Actually, I'm more interested in sfxocean performance test than the 3Dmark ones. Not everybody uses the Sager 9262 as gaming machines, they are power-house when it concerns CPU and Disks too.
I use mine as a virtualization platform to test multiple operating systems and management tools. So for me, having HDD performance test such as those given by sfxocean, is great.

Sfxocean, keep up the great work

 

You might want to monitor your system for read vs write patterns.

I use SAGER/Clevo's to build, demo and test enterprise integration applications. While I run DBs as part of the overall package, I've found that the HD profile is many more short reads than DB writes in our case. This comes from the way the vendors package their enterprise tools to run on XP with multiple enterprise components that are not at all optimized for the desktop and which collide in any end-to-end scenario execution.

When you get Multiple App servers, ESBs, transformation tools, Eclipse Dev tools etc. (in addition to the DBs) the response is terrible and costly to the overall fixed price project.

Based on our excellent experience with MTRON SSDs on an OPC desktop machine, I got an OPC built Clevo 901c based machine with MTRON SSDs. Expensive but far faster for our workload profile than our SAGER 9262 with multiple disks.

Your load may have similar characteristics.

Phil Schaadt

 

Thanks for jumping in here with the SSD info. I do not like having to plan and distributed my Oracle files across multiple drives, but that's the way it is when you want the best performance. RAID 0 is not equivalent for Oracle work, when the drives are mechanical imo (particularly in warehouse loading work with large bulk collect/read and writes).

I think SSD's will change all of this. When Intel gets prices down in 6 months I will be replacing the 3 WD mechanical disks with 3 SSDs in RAID 0. I hope to see performance that is equivalent to my current configuration for Oracle, and I can forget about manually distributing the database and index files across multiple physical drives.

That will have another advantage as well. All apps on the laptop, not just Oracle, will (possibly) benefit from the higher (realized) thruput, something I can not achieve currently with mechanical drives, unless where possible I manually distributed files that are accessed simultaneously by applications, across the disks

 

That is also our plan for all of our Dev boxes for our senior engineers. The productivity gain is worth it. The thing to watch out for is if the three SSD drives in a laptop will give linear scalability. In our workstation we went to an Areca server card because three MTRONs could not get to rated speed on the MOBO SATA (nvidia chipset). That is why I only went with two MTRON SSDs in the OPC (Clevo) laptop.

Thanks for the posts on your experiences. Perhaps we should start a developers/IT pros "lounge" thread here.

The big software companies are selling us all their enterprise tools for desktop development but if one of your developers tries to run a full product stack, all on a normal laptop, it grinds to a halt. We had a project last year where our developer could only hit "enter" 10 times in a day on a client supplied laptop because the huge software product stack was so slow.

Phil Schaadt

 

Got it, great point, I will remember that when the time comes...

 

Actually, I'm not sure if any of those apps would show much difference, and wouldn't really stress the system to highlight the sort of differences that sfxocean's testing brings out, mainly because, unless you're firing up every application you have at the same time, you're not really replicating the sort of parallel execution setting that sfxocean was testing.

That being said, you might still get the same sort of differences showing up if, say, you want to start up photoshop, internet browsing, a seriously big spreadsheet, all at the same time, and you're either sufficiently adept at multitasking, or sufficiently hopped up on caffeine (or other, not so legal stimulants) that you can keep each app busy doing something as you round-robin through each one.

Under that setting, I think that sfxocean's results could be extrapolated to support the conjecture that, if you had three drives in RAID, you'd get slower performance than if you had three drives operating independently, and had your apps installed in such a way that photoshop was primarily drawing from one drive, the internet browsing/surfing using a second drive, and the spreadsheet using the third, because in that instance, none of the applications would have to sit and wait for another to finish using the drive before it could use the drive (there'd still be queuing for the CPU and the FSB, but those are so much faster than the drives that the slowdown isn't noticeable in any event).

However, the hypothetical I came up with is just that - hypothetical - and there are probably other variables that would have to be taken into account before deciding what the best way is to distribute your apps and etc across multiple physical drives.

 

Note sfxocean hand distributed Oracle components across three discs. This is typical for tuning enterprise app performance. Even when installed as "development" tools on your laptop, these packages still have the underlying enterprise inter-component communications methods they are just executed within the laptop's OS. This often results in terrible performance as indicated above.

This is a different execution scenario than running specifically desktop or workstation apps under XP or Vista (Excel, PhotoShop, etc). The biggest positive impact for SSD is when your work combines demonstration enterprise Apps (say over WebEx) while using Excel and encrypt/de-encrypt to email documents to and from financial service clients.

The advantage for an SSD array is that tuning is not (as) necessary as hand tuning rotating drives as discussed above. That is assumimg the interface can handle the through put.

Similarly, in high performance media rendering workstations, server class “RAID” cards, like Areca, are used to give fine grain control across 8/16/etc. physical disks mapped to applications and activities.

Not typical gamer scenarios

Phil Schaadt

 

So where's the graph explaining the same with RAID 0?????

You only show performance with independant disks??