[Project] Tower of Power - Page 2 - Overclock.net

xlastshotx · 03-27-09

What kind of speeds do you see from that many hard drives? HDTach?

DJZeratul · 03-27-09

Quote:

Originally Posted by xlastshotx

What kind of speeds do you see from that many hard drives? HDTach?

This is what I am seeing right now on RAID5

which i have been told is low for what I should be seeing, but, its high enough for what i need, so i havent really investigated further.

as compared to 5 750GBs in RAID0 on ICH10R:

I think the higher burst speed has something to do with my QPI being so high or something.

Write speed on the RAID5 looks something like this:

T3h_Ch33z_Muncha · 03-27-09

two words: Holy Crap.

devsk · 03-28-09

write speed is low as well. I get about 150-160MB/s with just 3 drives in raid5 from Perc5/i. Try changing some parameters like readahead (RA with direct works best, ARA is slower), cache flush interval (4s is low, set to 10 or even 20 seconds), cached vs. direct (did not find cached mode to be faster) etc.

But seriously? 11 drives in total? you are nuts...

DJZeratul · 03-28-09

Quote:

Originally Posted by devsk

write speed is low as well. I get about 150-160MB/s with just 3 drives in raid5 from Perc5/i. Try changing some parameters like readahead (RA with direct works best, ARA is slower), cache flush interval (4s is low, set to 10 or even 20 seconds), cached vs. direct (did not find cached mode to be faster) etc.

But seriously? 11 drives in total? you are nuts...

13 actually, and I dumped the raptor for 5x750GB in RAID0 for my new system drive (1TB MBR for my system partition to be bootable, the rest as GPT for scratch)

Say, how do you change the cache flush interval?

devsk · 03-28-09

Quote:

Originally Posted by DJZeratul

Say, how do you change the cache flush interval?

I use linux mostly so you will need to find the megacli for windows. And then the command is:

Code:

MegaCli -AdpGetProp CacheFlushInterval -a0
MegaCli -AdpSetProp CacheFlushInterval -30 -a0

You will see a good gain in write performance with larger values. But too large and you are looking at potential loss of data. I set it to 30 like the command does above because that's my ext4 dirty expiry and commit interval as well. So, filesystem itself is not issuing any writes before 30s (well, unless you have so many writes that dirty_*ratio is passed before 30seconds).

This gets recorded in the firmware, so doesn't need to be set every time. But I do have it in my OS startup scripts.

DJZeratul · 04-04-09

Quote:

Originally Posted by devsk

I use linux mostly so you will need to find the megacli for windows. And then the command is:

Code:

MegaCli -AdpGetProp CacheFlushInterval -a0
MegaCli -AdpSetProp CacheFlushInterval -30 -a0

You will see a good gain in write performance with larger values. But too large and you are looking at potential loss of data. I set it to 30 like the command does above because that's my ext4 dirty expiry and commit interval as well. So, filesystem itself is not issuing any writes before 30s (well, unless you have so many writes that dirty_*ratio is passed before 30seconds).

This gets recorded in the firmware, so doesn't need to be set every time. But I do have it in my OS startup scripts.

cache flush of 15 seconds brings the write speed up to 350 MB/sec... SWEET.

Thanks for your help devsk!

It does however seem that Cached and Direct I/O do make a big difference, in Reads at least.

Read speeds with Always Read Ahead, Cached I/O

Always Read Ahead over Adaptive Read Ahead netted me about 2 MB/sec average, but my burst speed lowered significantly.

Read Speeds with Always Read Ahead, Direct I/O

Direct I/O just blows chunks on this guy. I'm not sure why, and I really don't know why it spikes near the end. But check out the difference in random access... 9.5ms vs 12.2ms...

linkinparkfan007 · 06-23-09

tell you something
that is one monster of a rig.

elite786 · 07-07-09

Hi,

i just joined especially to say thats a fantastic project, i have the perc5i on its way i wish id gone for the perc6i instead!

your thread is pure inspiration thank you!

i see your using the WD black edition drives i wanted the same, was going to get 5 to begin with, but reading some forums people where having issues a well known issue something to do with the drives always powering down and causing problems in any RAID!?

have you had any issues so far? if you havent ill be getting the same instead of the hitachi i was going to get originally.

thank you.

DJZeratul · 07-08-09

Quote:

Originally Posted by elite786

Hi,

i just joined especially to say thats a fantastic project, i have the perc5i on its way i wish id gone for the perc6i instead!

your thread is pure inspiration thank you!

i see your using the WD black edition drives i wanted the same, was going to get 5 to begin with, but reading some forums people where having issues a well known issue something to do with the drives always powering down and causing problems in any RAID!?

have you had any issues so far? if you havent ill be getting the same instead of the hitachi i was going to get originally.

thank you.

I haven't had any issues with drives powering down *knocks on wood*

But that issue seems to do with the TLER (Time Limited Error Recovery) "feature" on WD drives. There is a command line tool available that lets you correct this for when you are planning on using the drive in RAID. The WD RE drives have a higher TLER interval and you can set the regular consumer drives to be on par with this.

		Overclock.net - Overclocking.net > Components > Hard Drives & Storage
[Project] Tower of Power

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)