Page 1 of 2 1 2
Reply
Thread Tools
stefanmohl is offline stefanmohl
2013-08-19 , 13:45
Posts: 172 | Thanked: 170 times | Joined on Jan 2010 @ Sweden
#1
I finally got myself a Micro SD card for my N900, and it struck me, why not RAID-0 the Micro SD and the MMC. That should make it fast and big, and since I rsync everything each night, I can RAID-0 without fear.

Does anyone have any experience with doing this? Most importantly, does the latest Power Kernel have RAID-support?
 
pablocrossa is offline pablocrossa
2013-08-19 , 14:07
Posts: 189 | Thanked: 171 times | Joined on Jul 2011
#2
Originally Posted by stefanmohl View Post
I finally got myself a Micro SD card for my N900, and it struck me, why not RAID-0 the Micro SD and the MMC. That should make it fast and big, and since I rsync everything each night, I can RAID-0 without fear.

Does anyone have any experience with doing this? Most importantly, does the latest Power Kernel have RAID-support?
AFAIK (and I could be very, VERY, wrong) RAIDing works only with the same device and capacity i.e. a 64gb Plextor with another 64gb Plextor SSD, not a 32GB eMMC and a 16GB mSD (or even 32 for that matter)...

That being said it should be possible to spread loads by other means, just not RAIDing; I saw a kernel module (this was a couple of years ago) that simulated a Flash-Based device and every byte written to an odd-numbered sector was sent to one HDD and vice-versa, probably too CPU-intensive for N900 but who knows...
 
jackburton is offline jackburton
2013-08-19 , 14:13
Posts: 466 | Thanked: 661 times | Joined on Jan 2009
#3
Originally Posted by pablocrossa View Post
AFAIK (and I could be very, VERY, wrong) RAIDing works only with the same device and capacity i.e. a 64gb Plextor with another 64gb Plextor SSD, not a 32GB eMMC and a 16GB mSD (or even 32 for that matter)...
Wrong. RAID works on any block device. Doesn't matter what it is or how it's exposed.

If the kernel does support RAID, then it should be possible. I don't see how a RAID-0 will improve performance across the SD card and MMC.
The RAID performance will be tied to the lowest performing member of the set. If the SD card is slow, expect the RAID to perform slowly as well.

That being said. As a proof of concept, you could dd a few test images, mount them as loop devices, and then RAID them to see if it will work.
 

The Following User Says Thank You to jackburton For This Useful Post:
stefanmohl is offline stefanmohl
2013-08-19 , 15:33
Posts: 172 | Thanked: 170 times | Joined on Jan 2010 @ Sweden
#4
Hardware RAID often works best with identical disks, but the Linux software RAID is actually very forgiving, so I think the slight differences will be OK.

I am hoping that the striping will give me more bandwidth. Apparently there is an advantage to using an extra parallel swap partition on your sd card if the sd card is fast enough, so I was hoping that I could get the same advantage to disk access in general.

I made a large file from /dev/urandom and timed when copying it to both the MMC and the sd card (from root). The sd card and the MMC seem to have roughly the same speed, at least for sequential writes.

Does anyone have any experience with doing this? Are there any problems? Any advice?
 

The Following User Says Thank You to stefanmohl For This Useful Post:
jackburton is offline jackburton
2013-08-19 , 15:45
Posts: 466 | Thanked: 661 times | Joined on Jan 2009
#5
Originally Posted by stefanmohl View Post
Hardware RAID often works best with identical disks, but the Linux software RAID is actually very forgiving, so I think the slight differences will be OK.

I am hoping that the striping will give me more bandwidth. Apparently there is an advantage to using an extra parallel swap partition on your sd card if the sd card is fast enough, so I was hoping that I could get the same advantage to disk access in general.

I made a large file from /dev/urandom and timed when copying it to both the MMC and the sd card (from root). The sd card and the MMC seem to have roughly the same speed, at least for sequential writes.

Does anyone have any experience with doing this? Are there any problems? Any advice?
I think you should test with RAID'd dd'd images first and see how well it goes before going all in. How fast is the card? what kind of card is it? I'm curious on this as well even though I have no real use case for it...
 

The Following User Says Thank You to jackburton For This Useful Post:
Estel's Avatar
Estel is offline Estel
2013-08-19 , 16:56
Posts: 5,028 | Thanked: 8,613 times | Joined on Mar 2011
#6
You're doing it in pursuit of MOAR SPEED, eh? Pendrive/SD raids were tested numerous times, and performance is miserable. So miserable, that "flash raid" or "pendrive raid" has become some sort of pun-name*, when someone want to gain only numerological performance advantage (with no impact in real life).

/Estel

*See "Maemo administration" or "Hildon Board of Directors", for other popular pun-names.
__________________
N900's aluminum backcover / body replacement
-
N900's HDMI-Out
-
Camera cover MOD
-
Measure battery's real capacity on-device
-
TrueCrypt 7.1 | ereswap | bnf
-
Hardware's mods research is costly. To support my work, please consider donating. Thank You!
Last edited by Estel; 2013-08-19 at 19:30.
 

The Following User Says Thank You to Estel For This Useful Post:
stefanmohl is offline stefanmohl
2013-08-19 , 17:21
Posts: 172 | Thanked: 170 times | Joined on Jan 2010 @ Sweden
#7
jack: It's a a SanDisk 32GB UHS-1 30MB/s SD card.

Estel, yep, MOAR SPEED is what I am after! But I would prefer real speed, not just single digit percentage or extreme-case-only potential. However, I can't see that there is anything inherent in flash technology that prevents it from being RAIDed. AFAIK, one of the reasons for the high performance of SSD drives is that they have many flash-chips that are "RAIDed" internally.

You are mentioning pen drives, so maybe you are talking about some older measurements using the USB port? I have a reasonably good performance modern micro SD, so it should be significantly faster than your average pen drive...
 

The Following User Says Thank You to stefanmohl For This Useful Post:
Alecsandru's Avatar
Alecsandru is offline Alecsandru
2013-08-19 , 19:24
Posts: 439 | Thanked: 282 times | Joined on Oct 2012
#8
put away raid , i would like to see flashfire ssd driver on maemo , that would be something
__________________
N900 Beast featuring :
8GB OPTFS@ext4(*performance mount options)
overclock: cpu[125 - 1,150(*VDD1)+Dsp overclock
CSSU-T latest version
 
Estel's Avatar
Estel is offline Estel
2013-08-19 , 19:43
Posts: 5,028 | Thanked: 8,613 times | Joined on Mar 2011
#9
Originally Posted by stefanmohl View Post
Estel, yep, MOAR SPEED is what I am after! But I would prefer real speed, not just single digit percentage or extreme-case-only potential. However, I can't see that there is anything inherent in flash technology that prevents it from being RAIDed. AFAIK, one of the reasons for the high performance of SSD drives is that they have many flash-chips that are "RAIDed" internally.

You are mentioning pen drives, so maybe you are talking about some older measurements using the USB port? I have a reasonably good performance modern micro SD, so it should be significantly faster than your average pen drive...
Hm, In fact, I'm all for about MOAR SPEED myself Yes, I mean older tests done via pendrives through USB. Maybe it's indeed outdated, and I discarded idea too hastily? If yes, sorry for that, and lets look at it closer. The two (related) things, that make me think it won't work well:

1. On N900, using 2 swaps at the same time - one on eMMC and one on microSD, with equal priority - doesn't seem to help with speed, at all. In fact, it even gives *worse* results than swap on microSD *only*, due to I/O conflicts between /home and swap on eMMC (see point 2).

2. We have hardware controller on eMMC, that (transparently for OS - no control for it, at all) does some "nice" things, like forced hardware leveling. Now, it's well know, that flash memory doesn't have problems with multiple reads, but simultaneous multiple write attempts cuts performance by orders of magnitude (so-called I/O conflicts - not real "conflicts" in true terms, just I/O's getting in the way of each other, and slowing things down)

*if* you would be using software RAID on eMMC and microSD, you would have simultaneous I/O on both eMMC and microSD all the time, coupled with I/O for swap. IMO, it will lead to same performance penalties as in point 1. during normal usage, it can be avoided by having swap on microSD *only* (so I/O from opt, home, etc, doesn't get in a way with swap, and other way around, as they're two separate physical devices), but if you RAID them, there is no way around it.

/Estel
__________________
N900's aluminum backcover / body replacement
-
N900's HDMI-Out
-
Camera cover MOD
-
Measure battery's real capacity on-device
-
TrueCrypt 7.1 | ereswap | bnf
-
Hardware's mods research is costly. To support my work, please consider donating. Thank You!
 

The Following User Says Thank You to Estel For This Useful Post:
stefanmohl is offline stefanmohl
2013-08-20 , 00:27
Posts: 172 | Thanked: 170 times | Joined on Jan 2010 @ Sweden
#10
Hmm, according to the wiki page on balanced swap (same priority), it can be an advantage, no diff or slower, depending on the SD card performance, YMMV. For the wiki page author it was an advantage.

http://wiki.maemo.org/Swap_on_microSD

Have you tried using balanced swap on MMC and SD card on a modern SD card? I am asking just to try to reconcile your two opposing reports.
Last edited by stefanmohl; 2013-08-20 at 01:38.
 
Reply
Page 1 of 2 1 2

« Previous Thread | Next Thread »

 
Forum Jump


All times are GMT. The time now is 18:51.

Contact Us - Home - Archive - Top