• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

CF II Cards in Striped RAID???

statik213

Golden Member
Has anyone tried throwing 3~4 2GB CF II Cards thru IDE adapters in a striped RAID array? Should be interesting to see the performance, sustained transfer speed would be comparable to a regular HDD but what about access speeds? Wouldn't reading from Flah Memory have much lower latency than a mechanical hdd?
 
Interesting idea. I saw somewhere that some made a 4gb RAID 0 array with 4 iPod Shuffles...that was amusing. No performance figures thou.

Anyone got 4 large CF cards? I fancy trying to install windows on one.

Steg
 
Originally posted by: Steg55
Interesting idea. I saw somewhere that some made a 4gb RAID 0 array with 4 iPod Shuffles...that was amusing. No performance figures thou.

Anyone got 4 large CF cards? I fancy trying to install windows on one.

Steg

Got a link to the 4 iPod shuffle raid?
 
Nice idea - however, I think the raw transfer rates would be lower than those of a self-respecting hard drive. Anyway, the very low access time would rock 🙂
 
The best CompactFlash drives max out at around 10 MB/s. That's about four to five times slower than average IDE drives, and latest SCSI HDDs are almost ten times as fast.
 
Originally posted by: Peter
The best CompactFlash drives max out at around 10 MB/s. That's about four to five times slower than average IDE drives, and latest SCSI HDDs are almost ten times as fast.

qwll 4 x 10 mb/s is OK if it sustained
but I'm more interested in the Access Times... Shudn't they be a lot shorter than the fastest SCSI drives?
 
I tried "hard drive tach" on my ancient 32MB CF card and ancient Sandisk card reader. The results...a whopping 1.0MB/sec
 
Originally posted by: redly
I tried "hard drive tach" on my ancient 32MB CF card and ancient Sandisk card reader. The results...a whopping 1.0MB/sec

But aren't the new CFII cards rated at 70x where 1x = 1x CDROM => 70 * 160 kB/sec = 11200 kB/sec = 10.9375 MB/sec

Wouldn't four of these striped drives give at least 30 MB/sec sustained transfer?
 
... and still be half as fast as a decent single HDD? Besides, CF is PATA, with the drives supporting MWDMA2 mode at best and having zero (yes zero) cache. Since PATA does not do concurrent access, you don't get 20 MB/s from two 10 MB/s drives on the same IDE channel, you get 10.
 
....with much faster access times ??

I'm interested only in seeing the access times on these arrays... Shouldn't read/write latency be much lower than regular HDDs??? Wouldn't these drives be awesome as swap to put your swap files and OS on?
 
Zero cache won't be such a big problem on a "drive" with constant access speed - anyway, there is also cache on the operating system.
 
Just for the kicks I built a RAID-5 on three raw device which were on:
- a CF card on my mp3 player
- a file on a network-mounted filesystem
- a file on a memory filesystem

It was only done as a pissing contest to show people what software RAID can do, as opposed to hardware raid or raid provided only for specific controllers. As in "if one of your RAID-5 slices goes down, you can use any space you can scramble together, anything, anywhere, to hook it back into your RAID-5".

At the time Linux did not have a network filesystem interface to gmail, that would have been the final hack 😀
 
Originally posted by: MartinCracauer
Just for the kicks I built a RAID-5 on three raw device which were on:
- a CF card on my mp3 player
- a file on a network-mounted filesystem
- a file on a memory filesystem

It was only done as a pissing contest to show people what software RAID can do, as opposed to hardware raid or raid provided only for specific controllers. As in "if one of your RAID-5 slices goes down, you can use any space you can scramble together, anything, anywhere, to hook it back into your RAID-5".

At the time Linux did not have a network filesystem interface to gmail, that would have been the final hack 😀

Did it work?
 
Originally posted by: statik213
....with much faster access times ??

I'm interested only in seeing the access times on these arrays... Shouldn't read/write latency be much lower than regular HDDs??? Wouldn't these drives be awesome as swap to put your swap files and OS on?

What use is almost-zero access time when you don't get good throughput afterwards? The couple of milliseconds saved on the headstart will very soon be used up by the much much slower data transfer. Besides, write latency isn't even that good because a flash "sector" must first be erased before it can be written to.

Swap file on a FlashROM drive is particularly brilliant (NOT) because FlashROM cells have a limited number of erase/write cycles before they die. With the typical lifespan being 100,000 cycles, the heavy read/write activity on a swapfile will kill them within days.

If you want real zippy storage, you can use Quantum's RAMdrive devices - these are RAM arrays on a SCSI interface, with a real HDD as a nonvolatile backup. On recurring reads as well as writes, these have zero access latency and around 300 MB/s throughput. And if you want your machine to be faster although it's using the swapfile, then don't make the swapfile faster, add more RAM.
 
What use is almost-zero access time when you don't get good throughput afterwards? The couple of milliseconds saved on the headstart will very soon be used up by the much much slower data transfer. Besides, write latency isn't even that good because a flash "sector" must first be erased before it can be written to.

I did just want to note that some applications are almost ENTIRELY access-time limited (for instance, databases with a large number of very, very small records). If you're reading and writing 1K blocks of data all day, and all you care about is doing the maximum number of those operations per second, this would be WAY faster than a RAID of regular hard drives, because the hard drives will spend 90+% of their time seeking. Even desktop use tends to be dominated by seek/response time; how often are you loading 50+MB files (unless you're doing video/photo editing)?

A hard drive RAID might be a good order of magnitude (~100MBps versus ~10MBps) faster than straight flash-based storage. But access time for the flash storage will be *several* orders of magnitude better than on a hard drive (maybe .1ms or lower versus ~10ms).

Of course, you can't use standard flash memory as a swapfile, because (as noted) you'll kill it fairly quickly. You could, however, install programs on it, or use it to store frequently-accessed-but-infrequently-updated data. Flash-based drives that don't quickly die are available, but cost a ton. RAMDisks (with or without disk backup; you can get ones that just have battery backup as well) are MUCH cheaper (relatively speaking), and also have next to no latency and very high throughput.
 
Well, I didn't know of the limited write life-span of Flash memory so I guess that kills the idea.

Got links to the RAMdrives? I've know about them for some time but never got my hands on one.
 
Originally posted by: statik213
Originally posted by: Calin
Zero cache won't be such a big problem on a "drive" with constant access speed - anyway, there is also cache on the operating system.

???

The operating system will have disk buffers for both read and write operation, acting in almost the same way as a disk cache
 
Originally posted by: statik213
Well, I didn't know of the limited write life-span of Flash memory so I guess that kills the idea.

Got links to the RAMdrives? I've know about them for some time but never got my hands on one.

You might try a Google for "PuRAM" or "SSD" or "Solid State Disk". That should turn up a lot of info.
 
It might be an ok idea for a solid state silent computer that maybe just boots a small linux off the CF and runs in a RAM disk the rest of the time. You could do it even cheaper with something like knoppix on a CD, but then you have some moving parts. Coupled with a passively cooled cpu, that could be a nice system for basic internet use or for file serving if you don't mind adding some big IDE drives.
 
Originally posted by: Peter

Swap file on a FlashROM drive is particularly brilliant (NOT) because FlashROM cells have a limited number of erase/write cycles before they die. With the typical lifespan being 100,000 cycles, the heavy read/write activity on a swapfile will kill them within days.

Physically, the number of write/erase cycles is much less than 100k on modern flash cards. It's actually more in the range of 10k (for MLC, binary is a bit better). Steps are taken logically, however, to improve the lifetime of the card through wearleveling and other block managment techniques.

The transfer speed is being improved though. In the coming year, don't be surprised to see 66MB/s data transfer rates.
 
Back
Top