2 SSD's Raid 0 Outperform 4 SSD's Raid 0

[Jeff_michigan]

First, it appears that two (2) vertex 30gb ssd's in 4kb ntfs file size significantly outperform four (4) 30gb vertex ssd's in 4kb ntfs file size, but the opposite is true with larger cluster file sizes.

4 raid 0 vertex 30gb 4kb file size = read 145.5 MB/s; write 140.4 MB/s, vs.
2 raid 0 vertex 30gb 4kb file size = read 335.7 MB/s; write 317.6 MB/s.

but that the same raid 0 2-ssd's perform much slower in larger cluster sizes than when in raid 0 4-ssd's. See same links.

4 raid 0 vertex 30gb 128kb filesize=read 419.7 MB/s; write 569.6 MB/s, vs.
2 raid 0 vertex 30gb 128kb filesize=read 420.7 MB/s; write 322.7 MB/s.

test performe on ATTO benchmark software, links for reference below.

http://www.ocia.net/reviews/vertexraid/page4.shtml
http://www.ocztechnologyforum.com/forum/showthread.php?t=58002

Given this data is accepted as true, then if one wants to set up raid 0 with these vertex 30gb drives, he will get better MB/s performance with two ssd's raid 0 than with four ssd's raid 0.

Is this correct? Does everyone agree with this conclusion?

thanks.

jeff

 

[railman]

I would agree with test results. Once you go more than a 3 drive array then the results are minimal. This is mostly a limitation of the SATA/chipset interface.

 

[Jeff_michigan]

i understand what you are saying, and many on the web have the same opinion.

Respectfully, however, 4 30gb ssd's raid 0 128kb cluster size delivers 569 MB/s write and 419 MB/s read, compared to 2 of the same drives and cluster size in raid 0 which delivers 333 MB/s read and 317 MB/s write, the biggest gain is seen in write speed when moving from 2 to 4 ssd's striped. So, if the data is accurate, then there is a 253 MB/s gain in write speed moving from two to four ssd's striped, which is about a 125 MB/s gain per drive added.

My question, however, is this.

Is any effective storage space lost if the drives are striped at 128kb cluster size, compared to 4kb cluster size? In otherwords, for each file written that is not a full 128kb in size, a full 128kb of space is used, Yes? Thusly, if cluster size is at 4kb (not 128kb), then the effective space available is More since the overlap will be limited to 4kb instead of the unused portion of 128kb. Yes? Do you see my point? Thus, if this is true, then the effective available space will be more (even though speed is less) if one uses 4kb cluster size file system, instead of using 128kb cluster size file system, meaning that the amount of lost space is always minimized when using smaller cluster size file system. Does this logic ring true, or am i missing something?

My application is cad. I am perfectly happy with 2 drives at 300mb/s read and write 4kb size...my issue is getting the most usable space from the raid set, not the maximum speed...

Thanks for your prior response and thank you in advance if you elect to continue with the inquiry.

 

[railman]

Hmm, well your theory is correct but I believe that because of the way in which SSD drive controllers write data to the NAND cells that theory might not hold true. SSD drive controllers spread writes out over the NAND cells in a random manner not is a sequential order. This process is known as wear leveling and I think that it has a detrimental effect to your logic. It is my understanding that SSD's write data to cells much differently than HDD's. NAND cell blocks are 512kb in size, once data is written to a cell regardless of file size that cell will not be written to again until all cells on the drive have been written to and no free cells remain for writes or cells are marked as free by a TRIM command for example which, at present, does not work on a raid array so a consolidate free space defrag and then free space cleaner would need to be used.

Will you be using this array to host an OS plus apps and storage space?

 

[Jeff_michigan]

yes, the array will be used to host an os plus apps, with 1TB mechanical hdd holding images of 3 (more or less) OS & AP combinations, such that, when i want to do cad, for instance, I'll clone the OS with CAD aps to the array, however, when i want to do complicated (very large) large spreadsheets, i then would (after clean erase of the array - i forgot the name of the utility which you correctly refer to as space cleaner), then i would image the OS with Excell and other associated aps to the array... thus making the two 30gb array do "double duty", as it were...

 

[Jeff_michigan]

i simply can not afford larger ssd's at this time, but since i work on cad for weeks at a time, then work on spreadsheets for weeks at a time, i am willing to "retool" the array with the appropriate (needed) set of aps, from time to time...

 

[Jeff_michigan]

taking what you say about the ssd not re-writing a block until the disk is full, then my prior question is moot and my logic is flawed, since i will not actually fill the array before reloading another set of aps & os, to do another set for work for a few weeks...does that make sense?

 

[Jeff_michigan]

can you also set up os partitions on ssd array, separate from aps partition?

 

[railman]

Ok, you have 3 seperate OS, apps images on the 1TB HDD, can you boot to any one of these? Are you using Windows, what version?

 

[Jeff_michigan]

the idea is to have the os (win xp 64) and cad aps (autocadd and others) on a partition on the hdd spinpoint F3 partition, and to clone image this to the ssd raid array without any data files...

same holds true (but a different partition) for the os (again xp-64) and spreadsheet aps (excell and various add-ins) on a second hdd partition, that can also be cloned imaged to the raid array (as a clean overlay after the os and cad aps have beer removed e.g. utiltiy that brings the drive back to its original unwritten-to state)..

does this answer your question? i hope so, but if it's deficient, just let me know and i will elaborate...

 

[Jeff_michigan]

once the clone image is transferred to the ssd array, then i reboot with the spinpoint f3 hdd turned "off"...i can do this because i bought one of the front bay fan toggle switch sets (i bought the Lamptron 5 port military bay switch), ripped out the stock toggles (intended to control fans only DPDT switches), and replaced them with double pull single throw toggle switches (DPST), and control each hard drive with a toggle switch, thus allowing me to have certain hard drives on and keep others off, simultaneously...

 

[railman]

I think so. I am assuming you can boot to one of the OS apps images on the HDD, yes?

 

[Jeff_michigan]

once i save enough pennies, i'll replace the f3 spinpoint hdd with cheap jmicron ssd's to hold the custom os and aps images, since all i need those for is quick read to the primary array, thus not being burdened with the jmcron latency problem...at least, that's the plan at present...all powered by amd phenom II 545 mid range cpu with 6mb ondie cache...and 8gb ddr2...motherboard will hold 16gb, but without need for a ram drive, i can't see justifying the added 8gb (to total 16gb) since i don't think it will make that much difference for my purposes...

 

[Jeff_michigan]

yes, can boot to any of the hdd os, by hiding the partitions that i do not want to boot to...using say, partition magic or similar shareware utility

 

[Jeff_michigan]

since ssd are 2.5 inch, i can hold two of those per 3.5in bay, thus with 5 3.5in bay case (which is what i have), can hold 10 ssd's...at least that's the plan...

 

[Jeff_michigan]

as a side bar, i ripped out the stock 60mm amd heat sink and noisy fan, bought a stock amd 70mm heat sink (and noisy fan) $6, threw away that fan, added a 70mm to 120mm fan adapter ($8), and added a very very quiet 120mm artic pwm fan $15...since the cpu only uses 95watts and since i'm running it stock (no OC), i can't hear it at all...running an expensive 500watt seasonic power supply...so now the only noise in the case comes from the mechanical F3 hdd...oh yes, added passive cooler to ati hdd 4830 ver. 1gb vid card...so no fan noise there either...

 

[Jeff_michigan]

works great, just trying to get my arms around how best to configure ssd array for optimum performance...based on a very limited budet..

 

[Jeff_michigan]

one last config. tidbit...the mAtx motherboard i have delivers six sata ports, so i will add a pci-e x1 4 sata port card to hold the remainng 4 ssd's intended to be added to complete the system...the 4 sata port x1 card cost is $55..

 

[Jeff_michigan]

so, on balance, does anyone see any flaws with the architecture of the above described system?

 

[Jeff_michigan]

the last piece of the puzzle is that two cheap jmicron hdds will be placed in a 3.5in "raid cage" which are then ported to the SIL 3124 port multiplier card in the pci-e x1 slot, thus pushing read time to the max the x1 will handle, e.g. 250mb/s less overhead...this reduces available drives to 8 from 10, since each raid cage requires one direct sata cable to the port multiplier card in the x1 slot, but i can live with 8 drives...all pushing 250 MB/s read write speeds...and 420 mb/s read 320 mb/s write speeds on the primary 2 ssd raid 0 array, using the onboard amd controller connected to the onboard sata ports...

 

[Jeff_michigan]

for anyone interested, the case is a nzxt rogue case...picked that case because it is flat like a desktop case, but stands about 11 inches tall and 13 inches mol wide..so it has a flat surface that in effect extends desk top area...can stack books pads of paper etc. on top of it...motherboard tray, screwless of course...blue leds (did not like the red led version)..front door with key lock to keep kids off my system when i am not around..

 

[railman]

Ok, here's what I would do. When you wish to reimage the array use the Vista install disk and boot to it. Choose the recovery option at the install screen. You should be able to access a command prompt there. Open the command prompt and type DISKPART then LIST DISK. You will see a list of all installed disk on the PC. Next type SELECT DISK # where # corresponds to the number of the raid array from the list disk command. This will make that disk the focus of diskpart thus all subsequent commands will only be performed on that disk(s). Next type CLEAN ALL this will erase all data and partition information for the array. Now I have not done this on a raid array so I might be off here but I believe at this point you will have broken your raid array. Type LIST DISK again and check the list. If you now see 2 SSD's instead of 1 then the array has been broken. If that be the case then type CREATE VOLUME STRIPE SIZE=1024 DISK=#,# This will recreate your raid array.

Now exit out of the install and reboot to an OS on the HDD and then reimage the desired OS app install to the array. Now run AS Cleaner with useFF box checked on the array. This will reset all unoccupied cells to 1 which tells the drive controllers that those cells are free.

The end result should be an as new fresh raid 0 array with OS, app installed. This should be relatively quick, fairly easy, and yield excellent performance results.
Here's a link to DISKPART and its command options:
http://support.microsoft.com/kb/300415

Don't waste your time with the cheap raid adapter card as it will not perform the way you want it to nor would it support the use of SSD's. The only raid card that I am aware of in the market currently that supports SSD's is the LSi9260i series and sells in the $400.00 + range.

 

[Jeff_michigan]

thank you so much...this is the missing piece to the puzzle i've been looking for...
question: since im running xp-pro 64bit, should i still use the vista disk as you've outlined, or is there an appropriate xp pro 64 "command set" that will deliver the same results as you've outlined...the reason for my question is that i really like win xp 64 and i eventually will migrate to win 7, but i intend to skip vista and have so far been able to do that...

second point i have is that the x1 card i suggest using is not a raid card, but merely a pci-e x1 slot 4 port sata card that supports port multiplier...
it is a pci-e I/O controller with sil 3124 chip...that purports to accept
the intended raid cage (e.g. 2 ssd's raid 0 enclosure direct cabled to the I/O controller....
these are called mobile raid series ...an enclosure which holds 2 ssd's that is then mounted internally into a 3.5in bay...the raid 0 of these 2 ssd's is from the hardware contained within the enclosure...

anyway, that is the plan...i have already purchased them, but have yet to purchase the I/O x1 controller 4 port...by Syba model SY-PEX40008

so, in summary, even though i'm on xp 64bit, do you suggest using vista for the config...or is there an xp-64 bit protocal to replace the vista plan you kindly outlined above. and secondly,

i'm not adding a raid card to the x1 port, only adding a port multiplier card to connect raid cages to, each of whcih holds 2 ssds in a 3.5 in bay, each of which is raid 0...does that make sense,...did i explain it correctly...is there other information that i can provide to shed more light on the plan..

thanks in advance...

 

[Jeff_michigan]

the mobil raid series that i have (have purchased 3 of them) can be found at:

http://www.pacificgeek.com/product.asp?c=236&s=1295&ID=104485&P=F

the idea is to connect three of these, each containing 2 ssd's raid 0 array, to the pci-e x1 4 port card which has the SIL 3124 port multiplier chip on it...the manual for the raid enclosures states that they can be configured in this way, but it does not specify if 2 can be connected to the same card...we'll see...

 

[Jeff_michigan]

update: the internal raid 0 2 ssd enclosures have 1 cable per enclosure, carrying the raid 0 data, to the port multiplier pci-e x1 slot card...in earlier post this thread i said that each had two cables to the x1 card - that was incorrect and i edited the post above to correct the error....