Frequent system freezes + hd (ssd) optimization

I'm running UNE 10.04 on my AAO ZG5 with the 8GB SSD and 1.5GB of RAM. My only problem with this setup is a pretty significant one: the system freezes for anywhere between a couple to several seconds when doing just about any kind of task like switching tabs in a browser, or scrolling down a page (in a browser or pdf), or starting a program. In all cases, the HD activity light stays on for the duration of the freeze. I know the SSD is slow on these models, but I don't remember these freezes happening with the original linpus. I have disabled swap to minimize the use of the SSD. I have also followed several tutorials on optimizing the SSD performance but this hasn't solved the problem. Any ideas on how to eliminate these freezes? They are quite unbearable, but it is the only problem I have with my netbook!

 

replace SSD with 8Gb 300x (or 266x) CF ... and you don't expect "freeze" like the SSD Acer original solution.
performance on SSD is like a Class 8 SDHC card...poor, very poor. :|

viewtopic.php?f=43&t=59&start=380#p116203

 

I will seriously consider this mod since it will make the netbook much more comfortable to use, but in the meantime is there a way to at least avoid the freezing even if it doesn't fix the slow ssd speed problem?

 

freeze is a cache memory overflow on the SSD ... you can not avoid this (the OS must have an large and fast cache). :cry:
if the cache is overflow, the PC freeze but don't crash... :ugeek:

 

From what I understand, I will need a CF/ZIF converter with the right cable, and a CF card for this mod.

I believe this converter is appropriate: http://www.dealextreme.com/details.dx/sku.11814.

For the CF card, is there a point in getting one faster than 300X ? Is there a limit to the speeds the CF/ZIF converter and/or the interface can sustain? I see that there are some that are nearly twice as fast and relatively cheap. For instance: http://www.newegg.ca/Product/Produc...A-_-Memory+(Flash+Memory)-_-A-DATA-_-20211434.

Is there a way to tell which cards will work for the mod and which will get the best speeds?

 

I found that NOT using the elevator=noop option many sites recommend in fact increases performance significantly on the first One.
No idea why...

 

Since you have a lot of ram, try moving your /tmp files to a tmpfs (ie cache them in ram). This should save a bunch of read/writes from the SSD. There are directions on how to do so on many of the distro wiki's regarding the aspire one (ie ubuntu, arch, debian, etc.)

 

Are you sure you have the SSD version of the One?

Thanks, I had already done that.

In the end I decided to upgrade the SSD for the 32GB Supertalent SSD. The performance is MUCH better now. No more freeze-ups. It's a usable system now.

 

Unless there were some shipped with an 8 GB harddisk - yes.
I know it makes no sense whatsoever.

 

Well, not necessarily. The scheduler penalizes writes for reads. So, after the ssd performs badly on writes, and a scheduler prefers reads over writes, at least for special workloads a scheduler other then noop may yield better performance on ssds like those in the aao... This post gave me the idea to test with schedulers, especially anticipatory, which might be better than noop because of this

Klaus

 

Please let us know what you find. Having an idea of what schedulers work best for certain workloads should be beneficial to everyone, even to those of us who have upgraded the ssd.

 

These are results from the first test. Didn't have too much time testing it, but it seems interesting.

Code:

echo anticipatory >/sys/block/sda/queue/scheduler
echo 200 >/sys/block/sda/queue/iosched/read_expire
echo 2 >/sys/block/sda/queue/iosched/writes_starved

You can put this into /etc/rc.local, or for tests just do it on the shell (as root), which will be lost at reboot then, or you can reset everything with:

Code:

echo noop >/sys/block/sda/queue/scheduler

Klaus

 

What are you using for testing?

 

Mostly web browsing, pdf reading and such. Had no updates to install, this is a very good test, otherwise

 

It makes sense to test using normal daily activities but it would also be good to have some objective benchmarks to measure with.

 

Yeah, but this depends on what workload the respective objective benchmarks have.

the disk scheduler now is tuned markedly for favoring reads over writes. And it is the anticipatory scheduler, which describes its characteristics (you might check the web for diverse disk scheduler benchmarks...) It is true that more comprehensive tests would be better, but frankly, I do not have the time for that. I'll continue to try "my" typical workload, as 'Im doing it anyway If you start to use applications that write heavily, you're going bust anyway. The only thing is hoping that the actual scheduler parameter help get more reads into the queue than before. What might be interesting is checking whether this particulare scheduler instance e.g. streams films better (or at least not worse) than the noop scheduler.

One thing to consider, there are two different ssd models used, one slower with writes than the other. IIRC, I do have the "faster", so your mileage may vary.

Klaus