Internal storage options


@Exophase,

How important are random read speed results in the scope of things?

Also, can it indicate latency differences?
 
Last edited by a moderator:
@Exophase,

How important are random read speed results in the scope of things?
In the example I gave, unzipping big ROMs, random read speed doesn't really matter. It should be mainly big sequential reads getting the zip file into RAM then unzipping it, and then it's done. If the ROM is instead unzipped and not loaded fully into RAM random read performance could matter since it'll be scattered, but really the random access performance just has to be somewhat better than the original system was, which in DS's case isn't that hard. Even GBA emulation, which has almost no penalty for ROM random access, didn't really suffer from paging in big ROMs in 32KB chunks from the PSP's memory card (and I'm sure media performance there was nothing amazing)

A much higher performance game might have long loading times that are dependent on both sequential and random reads, if it's reading from a bunch of tiny files. Unlike emulated systems, that generally have little RAM, these may try to load a ton of stuff at once and make the user wait.

Random read performance could be an issue in general if your filesystem is very fragmented but AFAIK decent filesystems are supposed to inhibit fragmentation.
 
^ Thanks for that answer.

One further question, what would random read speed mean for the OS, is there cases where it may read different areas of the OS on storage during operation?

Or for example when its reading many files, like the thumbnail images for what could be many files on storage that you list.?
 
Last edited by a moderator:
vcoleiro: You are comparing eMMC bulk prices with MOQ 500 to µSD end-user prices with MOQ 1 with shipping often included. That is not really a fair comparison. If ED is going to ship the Pyra with a small but fast µSD card, he can also buy them at bulk prices. More importantly, he would need to buy eMMCs before PCB population can start, while he can buy the µSD cards later. The difference in timing also creates a price difference, because of interest losses and decreasing market prices.

The ODROID eMMC prices are interesting because they give a (rough) indication of what things could cost if we would aim for a modular eMMC approach. Of course they only give an upper bound on what the eMMC itself costs.

The benchmark link you gave (http://www.tomshardware.com/charts/microsd-cards-2014/-5-Random-Read-4-KB-QD-1-MB-s,3484.html) compares full-size SD cards, not microSD cards. The results don't necessarily transfer to microSD cards -- they could be better or worse, both are possible.

I agree with Exophase that benchmarks should be measuring comparable things -- it's annoying that you cannot find consistent numbers on the internet. Maybe if TrashyMG can do some benchmarks in a controlled environment (on an OMAP5432 dev board), it will give us better data.

I also agree that sequential read is the most important metric. Luckily this is also the metric that depends the least on the exact benchmark conditions.

I would postulate that besides sequential reading (which is the most important metric), sequential writing is the second most important metric. If people are going to use internal storage to store things besides the shipped base OS, they are at some point going to have to write them. Of course reading is way more frequent and important than writing, but fast write speeds are still nice to have.

My claim "microSD can be better than eMMC both in speed and in price" was based on the SanDisk Extreme Pro versus the ODROID eMMC module, since those two were the only ones for which I could find real speeds and prices. It's probably true though that in general it's "better speed or better price, pick one".
 
@ _wb_


Actually I gave both the bulk and single prices for that fast 16GB eMMC. $22 or $16.6 - actually cheaper at different places, ie singles can be gotten cheaper than $22. The point was that it is not the $39 that you quoted for the odroid module which came across as misleading. People would have interpreted that as eMMC being that price which it was far from. Generally at the same price points, eMMC is faster for read speed than uSD.


Also that link I provided , says it's for uSD. It's strange it also seems to include SD cards in it along with uSD. Not my link by the way, it was Grench who showed it for uSD card speeds. If it's wrong, that means all those that quoted uSD speeds from it have inadvertently used incorrect data. That would mean the only uSD speed test we would have is the storagereview link you provided which doesn't have the test criteria, which is next to useless for comparisons (see paragraph below)


Finally, I agree with you on the level benchmarking requirement. Even the seq read speeds being quoted are not comparable. For example the seq read speed on a 1000MB file will differ from the same test on a 4Kb file. Latency becomes a factor at reading smaller files. The 128MB/s eMMC seq read speed test was done on a 4KB file which latency would factor in on. The 95MB/s uSD seq read test was done on a 1000MB file which hides the latency. Ie You can't compare the numbers against each other.


I think EDs going to have fun testing both storage options . We really need to let him do it now. He should be able to do level playing field tests which everyone agrees is important. Otherwise we will just be quoting speed test figures for both eMMC and uSD which can't be compared because of the differing test criteria.
 
Last edited by a moderator:
Concerning effects of storage speed on a debian based system: the N900 has pretty slow internal storage (eMMC I believe) and it looks (no scientific test, only watching conky and top) like access to the large amount of small desktop files scattered all over storage is a major reason why running the catorize utility (an app menu structuring tool) or populating the desktop after boot takes really (like half a minute) long. If I look in my bin and lib folders the large majority of files is smaller than 64k. The slow N900 write speed actually throttles apt-get update, because it can't even keep up with a 200KB/s download speed (activating the development repository - 10MB compressed / 30 MB uncompressed package file - does test patience). 2G RAM on the Pyra are probably enough to mask most of this, but I think write speed and random read speed are important especially on the root fs. Speeds good enough for the Pandora are probably quite limiting on the Pyra. I don't care which technology is used, but it should be significantly faster on all performance metrics and not only sequential read speed.
 
Also that link I provided is for uSD, click on the link you posted, it clearly says so. It wasn't my link by the way, it was Grench who showed it for uSD card speeds.
No, those are full size SD cards. SanDisk Extreme Pro microSD doesn't even exist in 32GB, and Toshiba does not have Type 1 and Type 2 EXCERIA cards in its microSD range. If you click on the links in the rightmost column for the top SanDisk and Toshiba cards, you'll see that they are not microSD cards.

As for the prices: it is hard to compare prices, but I made an attempt by taking some recent historical market data from http://en.chinaflashmarket.com/pricecenter -- this source has numbers for both eMMC and microSD so things should be more or less comparable. I put that in a graph to make it easier to interpret. First the raw prices:

emmc-sd-prices.png


and then the prices per GB (which makes it easier to compare different capacities) :

emmc-sd-priceperGB.png


I think it is pretty clear from this data that µSD is in general cheaper than eMMC, and its prices are decreasing more rapidly than those of eMMC.
 
^No. We don't want the OS to take up the main cards.
I don't want to be limited in capacity by what seems to be good enough at T(production) -2 years..and once you make it replaceable might as well have it conveniently on the outside...whether you call it a main card, or internal storage is a matter of semantics, storage is storage

I believe this..in fact I think the most rational thing is to dispense with internal storage completely..it would be strange to have microSD and SD in the same device, why not go for 100% SD?
Why is it strange to have microSD and SD in the same device? I don't think that is any stranger than having, say, USB and micro-USB ports in the same device.
is any reason to have microSD other than if you are really pressed for space??

I think dropping fixed internal storage is a good way to lower the minimum sticker price without any big drawbacks. Consider the case of mobile phones: the flash storage is invariably either too small or there is a serious price premium for e.g. 64GB (which in SDcard land is not really 'big')

btw, why is microssd necessarily non-replacable?? couldn't you have something with on an internal or recessed eSATA or USB3.0??
 
@_wb_

As I said, that uSD link was provided by uSD supporters who have been using it to quote uSD speeds from.  If its wrong, why has it been used to quote speeds from? Additionaly , as I said, it just goes to show that we need to stop quoting numbers for speeds that dont have the same test criteria anyway and let ED do some real level playing field tests on both . I'm sure costs will come out of it also.

Also, in regards to 16GB uSD vs eMMC, I think its fair to say that from the real examples provided , rather than some random graph, that the prices for eMMC and and uSD are comparable - , however, eMMC always seems faster for the price point . Remember you need to factor in  the uSD cage price which that graph doesn't.    If you add in $3 for the cage , the prices match up for 16GB for example.
 
Last edited by a moderator:
Remember you need to factor in  the uSD cage price which that graph doesn't.    If you add in $3 for the cage , the prices line up. for 16GB for example.
But now you're comparing non-modular eMMC with modular uSD. If you want a modular eMMC, you'll need more than $3 (I think that was the price for the SIM+uSD combo cage btw, a uSD-only cage should be cheaper). And if you don't care about replacability, then solder the uSD card directly to the PCB without adding a cage.

Anyway, I think we can end this discussion now, since ED has announced that they'll let you choose yourself between eMMC and uSD. Pretty cool!
 
Anyway, I think we can end this discussion now, since ED has announced that they'll let you choose yourself between eMMC and uSD. Pretty cool!
Yeah that is an acceptable solution in my book. 
 
I wasn't talking about modular, its just a straight eMMC vs uSD( which needs a cage). Modular is a different kettle of fish.

Where has ED mentioned that?, that is cool !!   And yes, that sounds like an option that makes everyone happy :)
 
Last edited by a moderator:
Now we just need ED to give an official remark on the button labels and we can stop all the stupid arguing!

For now.
 
The two uSD cards that keep popping up are Toshiba's Exceria uSD and a SanDisk card of similar claims. I can never keep SanDisk's naming conventions straight.
I just ordered these: "Toshiba MicroSDHC Micro SD Card TF UHS-I 16GB Class10 Max 95M/s" and "SanDisk Extreme Pro 8 GB microSDHC Class 10 UHS-I 95MB/s" Both were between $20-$25 dollars.. although the SanDisk I had to get an 8GB instead of 16GB too keep my cost down. I figure after the testing the 16GB one can end up replacing the SD card in my phone and I'll use the 8GB on the dev board instead of the mystery card.
I finally got both SD cards in, one seemed to take the slow boat from China to get here.. I won't have time until next week to do any benchmarks as I'll be out of town over the weekend and I need to rebuild my image after corrupting it trying to get Debian Jessie working on it.
 
Last edited by a moderator:
Do a few write/erase cycles before benchs. Extreme Pro SD are usually incredibly fast during those first cycles.
 
Back
Top