If you prefer, please refer to my previous post at the following link for details.

http://www.webhostingtalk.com/showthread.php?s=&threadid=300102&highlight=Perc4%2FDi

PE2650 RAID5 (Read Cache enabled, Write Cache enabled) benchmark is 80M/75M
PE2450 RAID5 (Read Cache enabled, Write Cache disabled) benchmark is 35M/5M

My questions
1. Write Through = Write Cache Enabled? Write Back = Write Cache Disabled?

2. PE2450 doesn't have battery backup, is this the main reason why Write Cache is disabled by default and not changeable?

3. Since PE2450's write cache has been disabled, is this the main reason why Write is so slow?

4. Both Read Cache has been enabled on PE2450 and PE2650, why the difference is more than 100%? Does PE2650's CPU (Xeon) and RAM (DDR266) contribute so much in increasing the read speed on Raid5 comparing to PE2450's CPU (PIII) and RAM (PC-133)

5. Strange thing happened after I DISABLED the Read Cache in Array manager for PE2450 Perc3/Si, performance has increased to 70M from 35M. Why is this? Shouldn't be the other way round? (ie, Enable cache to have better and faster speed?)

6. Will 4 HDs benefits speed increase in Raid5 compares to 3 HDs?
But statistically HD failure increases with more disks right? How do we balance this?

7. What is your Raid5 benchmark on PE2650 or PE2450 or other vendor's raid? I knew IBM's Raid is particular slow, HP Proliant is ok. We use HD Tach, IOMeter and ATTO for testing.

Many thanks.

Question1:

Your best performance is with WB Cache and this is done when you set the policy for the drive (array in your case) to "Enable write caching on disk" and "Enable Advanced Performance".

This is the best definition I could find on WB and WT Caching.
Write-Back Cache: When the system writes to a memory location that is currently held in cache, it only writes the new information to the appropriate cache line. When the cache line is eventually needed for some other memory address, the changed data is "written back" to system memory. This type of cache provides better performance than a write-through cache, because it saves on (time-consuming) write cycles to memory.

Write-Through Cache: When the system writes to a memory location that is currently held in cache, it writes the new information both to the appropriate cache line and the memory location itself at the same time. This type of caching provides worse performance than write-back, but is simpler to implement and has the advantage of internal consistency, because the cache is never out of sync with the memory the way it is with a write-back cache.

Question 2
Good Ole Dell!!!! This is the reason why I quit using Dell servers. Yes, they have blocked enabling of write caching without the use of a battery backup on their proprietary MOBO's and Controller Cards. I could never find the registry hack to get around this either, but it CAN be done.

Question3
Not all PCI busses are the same. Dell machines with proprietary BIOS settings have always run slower than comparable SuperMicro systems using RAID controller cards. Also, not all cards are the same. I have found the dell supplied cards to be very slow performers.

Question 4 and 5
I have found ATTO tools to give some funny numbers where the write performance shoes to be higher than the read when caching is enabled. For this reason, I try to use IO Meter, it seems to give more "understandable" numbers.

Question 6
You should get better performance all the way around, but it is not dramatic. You should not worry about the consequences of additional disk failure, because the raid is redudant, and will continue to work until the disk is replaced. However, if you have 4 disks, and don't need the space of RAID 5, use RAID 10. It is far superior to RAID 5 in the write department.

Question 7
I have Atto and IO Meter benchmarks on any kind of array you can put 4 disks in for both SATA 7,200 and SATA 10,000 RPM. I will publish the data to the web when I have a chance convert the document from .xls.

Originally posted by hostingasia
If you prefer, please refer to my previous post at the following link for details.

http://www.webhostingtalk.com/showthread.php?s=&threadid=300102&highlight=Perc4%2FDi

PE2650 RAID5 (Read Cache enabled, Write Cache enabled) benchmark is 80M/75M
PE2450 RAID5 (Read Cache enabled, Write Cache disabled) benchmark is 35M/5M.....
have u actually asked Dell these questions? if yes, could they do anything about it?

could you yank the Dell RAID card (OEMed and somewhat propietrary) out and try maufacturer's supported Adaptec or LSI/AMI RAID cards? at least firmwares and drivers can be updated a lot more often than dell's OEMed controllers

Originally posted by cwl@apaqdigital
have u actually asked Dell these questions? if yes, could they do anything about it?


Good luck with that one! I spent nearly a month on the telephone with Dell on their Raid Issues. They will not give you any bench mark results. In my case I was getting 7 mbps write and they seemed to not be alarmed! Then I got this moron who said that the SATA channel could pump 150 mpbs and that I should be writing at least 140 mpbs. Go Figure!!!!

In most cases, when you talk to Dell, you will get a guy in India that will just read you the product documentation over the telephone.

I have also found that if you call around 1:00 - 3:00 PM Round Rock, TX time, you will not be routed to India, and the guys here in the US who are a little more knolegable.

Lastly, tech support never admited that there were any issues. When I applied for a RMA, finally, one of the customer support people understood what the heck I was talking about! I guess the return department sees these things coming back a lot.

Thank you so much.

never own a dell b4...do you have to 'prove' an actual hardware failure to Dell support in order to get the part replaced? can u get their 4-hour on-site replacement for 'performance' issues?

after all, it seems that you have to pay hundreds of dollars for the service plan built-in to server purchases, dell should also cover the performance issues?! or can you opt out the service plan and save a few hundreds? or I am just dreaming....

From my experience, Dell will generally work with you, expecially within the first 30 days of the return period. They will exchange the whole system or exchange components within that time frame.

Troubleshooting raid performance can be tricky. Usually, if the card boots and reads the bios, it is OK. I guess there could be a broken resistor or something that would be an exception to this rule.

Also, you could have a slow drive in the array. This would require I/O testing of each drive, which is a very time consuming process.

Also, your Mobo could have a slow PCI, or could be sharing with another device that is causing it to slow down.

3Ware has told me that generally proprietary boards (Dell, Compaq, etc...) usually perform slower than their OEM counterparts. I have also confirmed this with my own testing.

So if you are looking to get raw performance, I suggest an Intel, Supermicro ....etc board.

>do you have to 'prove' an actual hardware failure to Dell support in order to get the part replaced?

NO and you can get actual hareware replaced with any reason you want. :)

>can u get their 4-hour on-site replacement for 'performance' issues?

Yes.

>after all, it seems that you have to pay hundreds of dollars for the service plan built-in to server purchases, dell should also cover the performance issues?!

Yes, they do.

>or can you opt out the service plan and save a few hundreds? or I am just dreaming....

Yes, we opt for NBD and have our own backup hardwares like HD and PDUs.

>Also, you could have a slow drive in the array. This would require I/O testing of each drive, which is a very time consuming process.

Latest Maxtor Altas IV 10K shouldn't be slow.

>Also, your Mobo could have a slow PCI, or could be sharing with another device that is causing it to slow down.

Not really, it's build in Raid and 3 PCI slots are empty.

>3Ware has told me that generally proprietary boards (Dell, Compaq, etc...) usually perform slower than their OEM counterparts. I have also confirmed this with my own testing.

May or may not be true. Btw 3Ware's SATA raid only have 4M cache compares to Adaptec's default 64M and 3ware claims faster in performance. LOL

well, the lastest 3Ware 9500 series comes with 128M ECC cache standard and it can be upgraded to a whopping 1G cache

Originally posted by cwl@apaqdigital
well, the lastest 3Ware 9500 series comes with 128M ECC cache standard and it can be upgraded to a whopping 1G cache

They finally changed which is a great news.

Btw, who really needs 1G? :)

Originally posted by PSFServers
Good luck with that one! I spent nearly a month on the telephone with Dell on their Raid Issues. They will not give you any bench mark results. In my case I was getting 7 mbps write and they seemed to not be alarmed! Then I got this moron who said that the SATA channel could pump 150 mpbs and that I should be writing at least 140 mpbs. Go Figure!!!!


PSFServers,

Have you solved the problem at the end?

Yes, I agree with you whenever you mentioned something deeper into the technical spec. they won't know, but generally it's ok. In Asia, most of the tech. supports are very knowleageable and we liked their service much. Just remember don't ask them how to install things on linux/compile stuff, etc. :)

Oh...I've just called their tech support asking why my DracII console redirection graphics mode sometimes works, sometimes doesn't, and the answer is "oh, you are using a much latest firmware version than what we have here, and we don't have the updated manual (to read to you)" () I added those words in my mind. :)

Btw, Dell tech support told me not to trust any tool including ATTO, HD Tach or IO Meter. Best judge with the most basic testing tool. Get a counting watch and count the time.

So I did.

On PE2650 which is PERC3/Di, 700M took about 20 seconds, so it's about 35M, acceptable.

On PE2450 which is PERC3/Si, 700M took about 40 seconds, so
it's about 17.5M, which is ok.

I was using the SATA Controller from Dell, and it did not have a battery backup option, therefore, I could not enable write caching. I sent the hardware back for refund. They would not tell me how to hack the registry to get around the greyed out boxes, so it was gone.

Your PERC card has a battery option, so your only hope for reasonable performance through enabling write caching would be to add the battery.

My basic test is to take the Windows Server 2003 Operating system and copy it to the desk top. Then I time how long it takes to copy the file 540MB to the same location.

Here are some sample times.

4:24 Dell PE 700 - RAID5 Disk 4 (SATA 7,200)
1:59 PSF-R5D4 - RAID5 Disk 4 (SATA 7,200)
1:33 PSF-R5D4 - RAID5 Disk 4 (SATA 10,000)
0:48 PSF-R10D4 - RAID10 Disk 4 (SATA 10,000)

PSFServers,

Thanks for posting the benchmark.

Seemed SATA is far behind SCSI. Note even on the PE2450 which is PERC3/Si, 700M took about 40 seconds, so it's about 17.5M/s. On ration, 540m should take 31 seconds to complete.

Your data 0:48 PSF-R10D4 - RAID10 Disk 4 (SATA 10,000)
I noticed RAID10 took over 48 seconds to complete is a very long, so it's about 11.25M/s.

*P700 is P4 and DDR based. PE2450 is PIII and PC-133 based.

In any of the above case, I still think PERC Raid card is indeed very slow in write.

Update:

It's a testing software problem afterall. We upgrade HD Tach to 3.0 and PE2450 Perc3/Si speed is ok, about 50M/s