Various RAID levels and speed results [Archive]

Daynja

01-01-2005, 06:40 PM

I've wanted to go RAID for years, but I've never had the money. Well, I finally made it. I got four western digital 160GB hard drives with the 8MB cache. I chose to go with a cheaper "software" RAID card because I don't have THAT much money. I saw a card that supported RAID 5, and even if I didn't end up using it, it was nice to have and the card was only about $90. The card is a highpoint rocketRAID 454. I performed read and write tests using quickbench (and a different program, which nearly matched the results of quickbench shown here) with a single disk, RAID 0, RAID 10 and RAID 5. Latest BIOS and drivers were used. I'm not exactly pleased with the results, but they are what they are. Keep in mind that with RAID 0 the total size of the array is 600GB, RAID 10 is 300GB and RAID 5 is 450GB.

Sequential reads:
single disk: 46MB/sec
RAID 0: 41MB/sec
RAID 10: 40MB/sec
RAID 5: 41MB/sec

Sequential writes:
single disk: 33MB/sec
RAID 0: 82MB/sec
RAID 10: 41MB/sec
RAID 5: 14MB/sec

Perhaps I have set something wrong, and I really wish that is the case. I was expecting far superior read performance, but I didn't get that. It seems like all the reads are maxed out at 40MB/sec. The only real performance improvement over a single drive was with RAID 0, and that isn't an option for me. Slow write speed isn't really a problem to me, but I was hoping to get somewhere around 60MB/sec read speed.

Someone suggested to me that I try a different benchmarking program, such as HD Tach. I will do that soon. Is there anything else that I may be doing wrong? :confused:

Any comments or suggestions are welcome. Anyone else have any experience with the highpoint rocketRAID 454?

defakto

01-01-2005, 09:30 PM

If it's on a standard pci bus with other hardware you might be maxing out the bus.

Daynja

01-01-2005, 10:53 PM

If it's on a standard pci bus with other hardware you might be maxing out the bus.
I would consider that as a factor, except it is the only PCI card in the computer.

Motherboard is a Chaintech 9PJL Apogee (865 chipset)

Processor is a P4 2.8c

unhappy_mage

01-02-2005, 12:07 AM

I would consider that as a factor, except it is the only PCI card in the computer.

Motherboard is a Chaintech 9PJL Apogee (865 chipset)

Processor is a P4 2.8c
Well, the problem is that since all calculations for raid5 are done in software, that means the parity calculation stuff has to traverse the bus twice - once reading, once writing. This means that the bus bogs down faster.

Are you hip to trying linux for this? This type of thing may be faster with linux software raid than with the pci controller, especially since the onboard IDE controllers may not be routed thru PCI... that would be good.

What chunk size was used for testing?
http://www.mentallyretired.com/h/index.cfm/u_rogue_jedi (http://folding.extremeoverclocking.com/user_summary.php?s=&u=47426)

Daynja

01-02-2005, 12:38 AM

Well, the problem is that since all calculations for raid5 are done in software, that means the parity calculation stuff has to traverse the bus twice - once reading, once writing. This means that the bus bogs down faster.

Are you hip to trying linux for this? This type of thing may be faster with linux software raid than with the pci controller, especially since the onboard IDE controllers may not be routed thru PCI... that would be good.

What chunk size was used for testing?

The parity calculation slowdown for RAID 5 writes is apparent with the RAID 5 test, but that doesn't explain the slow read speeds of all the RAID levels.

I'm not doing linux for my main computer. :P

The stripe size for all the arrays was 64KB. For reading and writing the testing software reads and writes somewhere around 200MB. I also did random read tests and I have a graph of it somewhere. RAID 10 and RAID 5 showed slightly better random read performance than with a single disk. RAID 5 was slightly better though.

Ice Czar

01-02-2005, 01:57 AM

Someone suggested to me that I try a different benchmarking program,

IPEAK SPT yours for a measly $900 or so :p

http://www.hardforum.com/showthread.php?s=&postid=1026104985#post1026104985

Daynja

01-02-2005, 02:24 AM

hmmmm...looks like I will try winbench. At least it's better than what I used for the previous tests.

Ice Czar

01-02-2005, 02:41 AM

also give IOMeter a spin about as close as you'll get to IPEAK SPT
the locality issues are a flaw, so the results are more indicative then definative

Winbench is good for STR
check out and translate Femme Takkens SATA Benchmarking
he also employs IPEAK SPT like Storage Review does

edday

01-02-2005, 06:20 AM

Ice Czar

01-02-2005, 11:58 AM

youll find very few RAID cards that dont have a substantial penalty in write performance in RAID5, the parity calculations take time

I run an IDE RAID5, with one of the few cards with its own dedicated XOR RISC Processor
and write performance isnt all that bad, provided your talking about a single user percieved delay, but Its not large accesses, simply storage

If you actually need redundant performance space, look into the NetCell SyncRAID
which is a trick implementation of RAID3, its a special case however as its specifically for large file writes, such as youd see in video editing, it would actually be a worse performer in small database accesses

http://www.netcell.com/

specifically read > http://www.netcell.com/technology%20info/SyncRAID%20Removing%20RAID%20Adapter%20Bottlenecks.pdf

if you havent already you should probably read this first for some context
http://www.storagereview.com/guide2000/ref/hdd/perf/raid/index.html
(the whole subsection)

defakto

01-02-2005, 12:36 PM

Just because it's the only pci card you have doesn't mean it's the only thing on the bus. Alot of onboard sound and nics, as well as your onboard ide channels usually, depending on teh age of the board and what chipset, share the pci bus.

Daynja

01-02-2005, 01:07 PM

I'm looking into getting the same card myself, and upgrade my raid1 array to raid5. But one thing i'm worried about is the cpu usage of the card in raid5 mode. Did you happen to see what kind of cpu usage you were getting?

Also a while back i did read that this card did have rather poor write performance in raid5. I'll try dig up the article later and see if it says why.

I didn't know about this card when I bought it, but for about $50 more, you can get the highpoint rocketRAID 464, which has some processor for XOR or whatever the hell it is. If you were planning on going to RAID5, that's probably the best way to go, unless you want to spend $300+ for a really nice RAID5 card.

CPU usage wasn't that much with software RAID5. Yes, there were spikes to 50% usage, but for the most part it stayed under 10%, but yes, that was also a lot more than any of the other RAID levels or a single disk.

About the PCI bus, it's the 865 chipset, so I don't know what's on it. The onboard NIC and onboard sound are probably on it.

Ice Czar

01-02-2005, 02:31 PM

The onboard NIC and onboard sound are probably on it.

make that definately
as well as the IDE Channels and onboard RAID Controller if enabled, the COM ports, USB, and Firewire if applicable

review > PIRQ Routing Tables (http://www.sudhian.com/showfaqs.cfm?fid=7&fcid=26)
happens to be for one of my old boards, but its about as good an explaination of PCI resources as your likely to find

my workstation in the sig is an atypical animal
with 4 main buses, the AGP Tuinnel, dual PCI-X Buses
and a legacy 32bit 33MHz PCI bus
that has all the USB, Firewire, SATA RAID Controller, and IDE Channels on it

one of the PCI-X buses has the onboard gigabit NIC

PIRQ Routing tables are commonly published for servers and workstations, but harder to find for desktops ;)

defakto

01-02-2005, 03:18 PM

tyan thunder k8w?

Daynja

01-02-2005, 03:26 PM

Would changing which PCI slot I used make any difference? I used the second one down because my video card gets really fricken hot.

Ice Czar

01-02-2005, 04:48 PM

tyan thunder k8w?
indeed with the AMD 8151, 8131, and 8111 Hypertransport chipset
definately not "typical"

Would changing which PCI slot I used make any difference? I used the second one down because my video card gets really fricken hot.

yup

to what effect is unknown of course, youd need to know what else is sharing the PIRQ
ideally youd keep devices that are busy paired with devices that arent

for instance putting it on wiith a NIC if its a server wouldnt be all that good an idea since both would be constanly accessed, so pairing it up with a USB device would be better

I use my KR7A as a cheap seats NAS, (its just me so it doesnt need much)
thus my Promise SX6000 is on slot 6 w\ PIRQ 3
and there is nothing in slot 4, and I only have a Lexar Jumpshot Flash reader for USB
but I might be changing that soon

PIRQ_0 * AGP card and PCI Slot 1
PIRQ_1 * PCI Slot 2 and PCI Slot 3 and ACPI
PIRQ_2 * PCI Slot 5 and RAID controller (if present and enabled)
PIRQ_3 * PCI Slot 4 and PCI Slot 6 and USB

defakto

01-03-2005, 12:40 AM

Yeah I love my k8w. Best board i've bought in a while.

Daynja

01-04-2005, 10:20 PM

Well I tried it again with HD Tach, and for RAID 0 it averaged 60MB/sec read and for RAID 10 it averaged 46MB/sec read.

I've been trying to test RAID 5, but the windows software crashes the computer while it's trying to build it. So I'll have to have it do the 2 hour thing in the BIOS while I sleep tonight.

twistacatz

01-05-2005, 11:11 AM

Daynja,

I have the same card which I use in raid 5. My read speeds are at 86MB/s but my write speeds are a sad 8MB/s. I'm using 7 200GB seagates. I plan to switch to a 3ware card in the upcoming weeks. I'm also planig to switch to a MB with 64bit slots. Let me know how things work out for you.

Daynja

01-05-2005, 04:55 PM

Write speeds with RAID 5 will always be slow, even if you have a "hardware" RAID card. But 8MB/sec is pretty slow. I haven't tried a program that accurately measures write speed yet though. The "false" tests above showed 14MB/sec for RAID5. But the write speeds seem to be correct. The read speeds are way off though.

I just completed a test with RAID5 and HDTach...interesting results. Why does it spike up and down like that? Also notice CPU use. I know it's mostly a write CPU hit (which isn't measured here) but the CPU is hardly used during reads, even with software RAID5.

http://www.probe-gt.com/misc/raid5-hdtach.gif

Daynja

01-06-2005, 02:46 PM

Here is a comparison with a similar RAID5 array, except that the red line is a hardware RAID 5 card, and the blue line is my software card.

The red line is 4x maxtor 200GB hard drives with a 3ware hardware RAID 5 card.

The blue line is 4x WD 160GB hard drives with a highpoint rocketRAID 454 card.

http://www.probe-gt.com/misc/compare-raid5.gif

Is the difference really that much? I would think the hardware card would be faster, but not like this.

Since I'm kind of disapointed with the results of my RAID setup, I was thinking about spending even more money to get something decent. I've already spent over $500 to get something that's only slightly faster than a single drive. At least I have the redundancy...and the computer crashing... :rolleyes:

What are some opinions on this card:

http://www.newegg.com/app/ViewProductDesc.asp?description=16-102-036&depa=0

It's the only hardware RAID5 card that I could see myself buying.

twistacatz

01-06-2005, 03:02 PM

lol, wow trust me I feel your pain, I'm in the same exact situation. I get the same read speeds that you did which are not bad at all and . But its the write speeds that are killling me. I already decided to ditch the 454. If I was you I would not get that Promise card becuase really its not that much better then the 454 from the benchmarks that I have seen. I went with the 7506-8, as I believe you should. Everyone always recomends it and it seems to be the standard. It's also the fastest. I know it cost a little bit more, but like you I just spent a lot of money so I'm not going to let the price get to me... Plus its not that much more expensive. The promise off of newegg is $181.00 while I paid for a used 7506-8 off of ebay for $257, a little more cash but I would say its well worth it. I think you'll change your mind when you look at this:

http://www.3ware.com/products/pdf/Benchmark_7500-8_RAID-5.pdf

Daynja

01-06-2005, 03:39 PM

Why not the 3ware 7506-4 which looks like the same, but only 4 IDE channels, which is all I'd really need.

http://www.newegg.com/app/ViewProductDesc.asp?description=16-116-019&depa=0

btw, I went looking on ebay, and all I see is new hardware, and it's all more expensive than newegg. Check out this one...LOL

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&category=39968&item=5150857957&rd=1&ssPageName=WD1V

I don't really care about the slow write speed. I was just hoping to get reads that were 60-80MB/sec

twistacatz

01-06-2005, 03:58 PM

You know I looked at your results incorrectly, 52MB/s is very slow! I was able to get at least 80MB/s, I think almost 90MB/s. You should be able to get that with the 454. But yea if you are only using 4 drives look for the 7500-4, or the 7506-4. Those would be the best, but there are even some older cards that you can find if you search around ebay for "3ware" that would still work out awesome for you.

Your right there are no used ones right now, but wait a couple of days or even weeks and I'm sure someone well sell one, thats whats I had to do.

BTW that ebay listing you pulled up was for 10 cards not one, thats why it was so expensive.

Daynja

01-06-2005, 04:04 PM

How many drives did you have to get 80-90MB/sec ?

I think I'm just gunna stick with this card...I really can't spend any more money on this. If anyone has any suggestions as to why the speed just isn't there, let me know. Using RAID 10 isn't any faster and I lose 150GB of storage, so I think I'll just stay with RAID5, even with the slow writes.

btw, I did try the card in another PCI slot, and that didn't make any difference.

twistacatz

01-06-2005, 04:38 PM

Daynja

01-06-2005, 04:51 PM

My raid consisted of 7 drives, which could make the difference. When I upgrade maybe I can bench 4 drives for you and tell what kind of speeds I get. :) To make up some of the money you could return the 454 or ebay it, I think the 3ware is what you should do if you need the speed, 14MB/s Writes/ 52MB/s reads are just way to slow. I know its more money but if you can do it, and you have a need for speed, I would.

I don't NEED the speed...I just want it. :)

I probably couldn't sell the card for more than $60, and the 3ware card is like $300, so I think I'll just stay unhappy for now.

I think with 7 drives that would probably make up the difference between 50 and 80MB/sec.

I also have two 40GB hard drives laying around (with no computer for them) I know I could put these into a RAID 0 array on the same controller for speed only. How much would the speed of the 4 drive array be affected by this? I was thinking a little and there are some situations where I might need a fast write. I think I'll go try them on the slow RAID5 now and see if that's even fast enough.

BTW, what kind of case/power supply are you guys using with 7 drives? I had to buy a new case and power supply just for this. I also have 3 optical drives so it's not easy.

twistacatz

01-06-2005, 05:28 PM

I think you could sell your card for more like $80, and I'm sure you can buy a 3ware card for under $200. We are talking about $100 give or take. Having raid 0 on the card would do nouthing to the raid 5 array becuase no calcuations have to made for raid 0. I am using a Antec true 430W. This is what my server case looks like:

http://1stclasspcs.com/uploaded/compare.jpg

Daynja

01-06-2005, 05:43 PM

LOL....I have 3 of those cases. In fact, that's the case I just replaced. I really like that case, but there's barely any room for hard drives. I use all 4 5.25" bays for other things.

About adding drives, I was talking about having 2 drives on the same IDE channel, which I don't do currently because there are 4 channels and I have 4 drives. Doesn't having 2 drives on the same channel reduce performance?

I've just recently switched it to RAID 1/0 because this highpoint card is worthless and can't do RAID5 like it says it can.

So in short: NEVER BUY HIGHPOINT. :)

twistacatz

01-06-2005, 05:57 PM

I have 2 drives on the same IDE channel with raid 5 and it works fine. Yea that case is old school, it was the first PC case I bought back in 01.

Daynja

01-07-2005, 12:47 AM

Are you running the 3ware 7506 right now?

Do you have the 64-bit PCI slots? I don't, but I guess it's not that much of an issue?

What do you think of the 3ware 7500? I know it's older, but it's a lot cheaper. What is the difference? Will it work with drives larger than 137GB ?

twistacatz

01-07-2005, 09:02 AM

Are you running the 3ware 7506 right now?

I am not running the 7506 yet, I have to wait untill I can move all my data off of the current raid. So I should have it up and running in the next couple of weeks.

Do you have the 64-bit PCI slots? I don't, but I guess it's not that much of an issue?

Well I had to learn abuot the 64-bit slots after I got the stupid 454. You don't have to have them and your going to get much better results with this card over the 454 regardless, but if you want to see the 3ware reach its full capability you have to use a 64-bit slot. And this is why: the 32-bit slots are limited to 133MB/s for everything on that bus, that includes Lan HD's everything. So if anything else is using bandwidth on the bus its taking away from what the 3ware card could be using. I want this card to go as fast as it can go so I am buying a MB that has the 64-bit slots which have there own bandwidth. To bad all of the boards are $100+ and MP, but if thats what its gonna take, I am going to upgarde to it. For you I wouldent stress it.

What do you think of the 3ware 7500? I know it's older, but it's a lot cheaper. What is the difference? Will it work with drives larger than 137GB ?

The 7500 would be awesome for you, it definitely can see large hard drives and the only diffrence between the 7500 and the 7506 is that the 7500 can run 64-bit/33mhz while the 7506 can run 64-bit/66mhz, which would make no diffrence unless you had a 64-bt slot.

Daynja

01-07-2005, 02:34 PM

The 7500 would be awesome for you, it definitely can see large hard drives and the only diffrence between the 7500 and the 7506 is that the 7500 can run 64-bit/33mhz while the 7506 can run 64-bit/66mhz, which would make no diffrence unless you had a 64-bt slot.

I finally found somewhere on 3wares website that says the 7506 offers a 25% improvement over the 7500. Now that you mention this, I suppose that only applies to people using the 64-bit slot?

You're right, I did take a look at 64-bit slot mobo's and there is no way I'll be getting one of those anytime soon.

So whats the deal with pci-express? That's on a lot of new motherboards. I see they have a 16x slot for video cards, and this tiny 1x slot for everything else. Is the tiny slot good enough for raid cards? Or do they expect us to not use a video card?

twistacatz

01-07-2005, 03:29 PM

I mean this card works in a regular PCI slot, so I wouldent worry about pci-express.

Daynja

01-07-2005, 03:57 PM

I mean this card works in a regular PCI slot, so I wouldent worry about pci-express.

I was just wondering, I know this card has nothing to do with pci-express.

btw, I am returning the highpoint rocketraid 454 to newegg.

Lazn_Work

01-07-2005, 04:15 PM

I mean this card works in a regular PCI slot, so I wouldent worry about pci-express.

PCI-X is not PCI-Express.

==>Lazn

twistacatz

01-07-2005, 04:17 PM

PCI-X is not PCI-Express.

Dude I know that, Daynja was asking about PCI- Express.

PCI-Express is cool but first off there are no AMD boards that have it yet that are out. Second there are few things out for it in the raid card market, in fact raidcore is the only one that comes to mind.

But when hardware besides video cards come out for PCI-Express it well be great becuase each slot has 200MB/s of bandwidth.

PCI-X is awesome with raid becuase bandwidth can peak at 528MB/s when using a 64-bit/66mhz slot.

Daynja

01-10-2005, 07:25 PM

Daynja

01-11-2005, 12:07 AM

After reading through this extensive review, I'm thinking about getting the promise fast track SX4060 or SX4000.

http://www.tech-report.com/reviews/2002q4/ideraid/index.x?pg=1

The 3ware card is still very good in that review, but the promise seems a little better. Also I could get a 256MB cache pretty cheaply.

If anyone knows the difference between the SX4000 and the SX4060, I would like to know. As always, it's hard to find the difference.

Ice Czar

01-11-2005, 12:33 AM

LOL

did you actually follow my first link?
(if not go back and then access the next to last link in that post, which is Femme Takken's
SATA RAID roundup youll have to translate it as its in Dutch)

he employs IPEAK SPT
if you dont feel like translating it the benchmarks are in English and start to get serious on page 8

:p

Ice Czar

01-11-2005, 12:47 AM

the SX4060 is a 32-Bit/33 or 66 MHz PCI 2.2 interface as is the original SX4000
the SX4030 is a 32-Bit/33MHz PCI 2.2 interface as is the SX4000 Lite

the original SX4000 supports 256MB of SDRAM
the SX6040 supports 64MB of ECC SDRAM

http://www.promise.com/product/product_detail_eng.asp?productId=94&familyId=2
http://www.promise.com/product/product_detail_eng.asp?productId=122&familyId=2

Daynja

01-11-2005, 02:31 AM

ok so is your point that the promise controllers are good? Those benchmarks seemed to be pretty good for the promise controller.

Is the only difference between the SX4000 and the SX4060 that the SX4060 comes with 64MB of ECC RAM, while the SX4000 doesn't come with any RAM, but requires you to buy at least 64MB ?

Is there any reason to buy the SX4060 instead of the SX4000? I have no problem buying a stick of RAM...I might even get more than 64MB.

Although ECC RAM isn't required, is it a "good idea" because it's dealing with data, or is it not neccessary? It's a lot more money. :(

Oh man here come the questions...Even now I'm still debating over RAID 10 and RAID 5. I already have four 160GB drives so that's not an issue. I really just want the fastest speeds. The difference between 300GB and 450GB doesn't matter to me.

twistacatz

01-11-2005, 11:25 AM

I'm telling you, don't skimp out like you did with the highpoint (your going to end up with another return), drop the ends and get the 3ware. Many of those benchmarks on that Dutch site are not really relative to what your going to be doing. The only stat that you need to know is throughput, and your going to get better write/reed speeds with the 3ware along with a more polished feature set. Lets not forget the XOR proc helps a lot if the raid arary is going to be your main computer, because your system can get held down big time when transferring files with the promise.