Opposite of improving disk performance

[Tsaukpaetra]

I have an outright bizarre request: How do I REDUCE disk performance?

I decided that I probably have a bad IDE-Sata converter, but since I can't immediately replace it I need to fudge with it until I can.
The problem I'm facing is that intensive IO tends to overload the controller into an unrecoverable lockup, which eventually causes very bad things to happen. Since this particular pool is not normally under high load, it's not usually a big issue, but when it happens... yeah.

If I can reduce the bandwidth the controller is running at, I hope to minimize this problem for the mean time, by reducing the IO rate hopefully the controller won't get saturated and die.

Are there any sysctl variables or anything I can do to do this?

 

[cyberjock]

I've never heard of anything that you could do to deliberately reduce disk performance.

And I just have to say, what you are planning seems like a really bad decision. I'd probably shutdown the server until you can get a replacement. Not to mention IDE to SATA converters have always been a bad idea.. particularly for servers.

 

[toadman]

I would think you'd have to write a custom driver to throttle the I/O. I think you're better off just not using the server until you get a proper controller.

 

[Tsaukpaetra]

And I just have to say, what you are planning seems like a really bad decision.

I know, it really breaks me up to even consider asking such a thing (it sounds retarded in so many ways), but hopefully I'll have something workable by next week. It's just been an ongoing problem that prompted me to think of solutions.

The weird thing is that it's fine until I start putting load on it. For example, I can order a scrub and watch it complete fine, but as soon as I try upgrading ports the server will crash. The sad part is that I'm mitigating the effects of the instability by rsyncing to a Windows machine :/

I can say with high confidence my setup is terrible, but this is why I've been upgrading! I just got rid of my trusty NSLU2 last month, so things are slowly getting better.

Luckily I seem to be the only one who will be using the server to any appreciable degree, so my personal abstinence will probably be enough for now. ;)

 

[cyberjock]

Scrubs are about as intensive of a work as possible. To be frank, if a scrub is completing fine but you can't upgrade ports then I don't think your converter is the problem. What's your hardware specs, FreeNAS version, etc?

 

[Tsaukpaetra]

Laughable I suppose, but better than I had before.

Consumer Laptop IBM T42
1.7 Ghz Single Core x86 Processor, No PAE
512 Mb DDR2 Ram (502 Available to OS)
Intel Gigabit Ethernet

Pools configuration:
Internal:
1 Tb Sata 300 disk connected to IDE-Sata connector on Secondary IDE channel

External:
2 Tb Sata 300 disk connected to USB Enclosure to USB 2.0 Port 2

Boot drive is Sandisk 16 Gb thumbnail flash drive on USB 2.0 Port 1

 

[Tsaukpaetra]

For what it's worth, I'm going to replace the system with an Acer Aspire EasyStore H340, making this the first time I've had a 64-bit server. Yay! :D
I wonder though, since this is a 64-bit machine, will my existing jails will be completely useless since I'm moving from 32-bit to 64-bit? Probably a question for another thread if I have issues I suppose...

 

[krikboh]

That Acer system is not appropriate for FreeNAS either.

 

[cyberjock]

No.. your problem is that your machine is disgustingly outside of anything resembling a FreeNAS box's minimum. 8GB is the minimum RAM for FreeNAS. If you have less you should just *expect* problems. Period. To be honest, I wouldn't even consider messing with your server until you read the hardware requirements in the manual, then upgrade to 8GB of RAM per the stern warnings in the manual.

 

[Tsaukpaetra]

I know I know, but I just haven't had time to throw down enough money for that. My current setup would be fine were it not for the failing IDE adapter.
I saw the acer device and thought, "64 bit, four times the memory I currently have, seems like a good step up from my current machine", but if 8Gb is required, I'll have to wait a while longer for a true and proper upgrade.

 

[cyberjock]

Well, I'll just say "I hope you have backups". Typically when people have <8GB of RAM they have problems. Many have had a single server crash that resulted in a corrupted an unmountable pool.. aka total data loss.

 

[Tsaukpaetra]

Yup, mentioned before, it's going to a Windows machine via rSync. The only thing I don't have a good backup for is the jails, which were put on the internal drive for speed (peh).

I'll tell you, the network in my house and all of the devices on it are scary, as in post-apocalyptic scary. Most of it is made of undead machines franken-built to kinda-work. One day, perhaps when I win the lottery or something, I can finally replace these 2000's machines with new hardware. ;)

 

[Tsaukpaetra]

For now though, I'm going to try an even stupider idea, using powerd to set the CPU to the lowest frequency, thereby starving all processes of any speed whatsoever, giving the disk no chance to experience saturated IO (at least in theory). At 100 Mhz, I'm going to run the portsnap -a command overnight, we'll see how long it survives....

 

[cyberjock]

That idea had crossed my mind, but I'm not aware of how you could force the CPU to a lower frequency even under load. Also, depending on the CPU, powerd can cause performance(hey.. who cares for your situation) and stability(you'll probably care because you want stability) and for many CPUs it actually doesn't change frequency at all despite it claiming to. Powerd sends the request to the CPU, but the CPU doesn't acknowledge the commands nor obeys them. But hey, you got nothing to lose but your data, right? ;)

 

[toadman]

Not sure what software you'd use, but it can be done depending on the system in question.

Another use case is laptops. Obviously not applicable here, but just an example that it can be done. http://forum.notebookreview.com/har...arket-upgrades/531329-throttlestop-guide.html

 

[Tsaukpaetra]

Yeah, powerd is definitely working, I can set the frequency to a variety of speeds:

Code:

dev.cpu.0.freq_levels: 1733/27000 1516/23625 1333/21342 1166/18674 1066/17571 932/15374 800/13800 700/12075 600/10350 500/8625 400/6900 300/5175 200/3450 100/1725

and in the course of 11 hours it hasn't managed to finish compiling perl5-5.16.3_6, BUT! it hasn't crashed either
System load is at ~5.15, which makes sense I think.
With this much promise, I think I'll bump it to 400 or maybe 800 so updating the 104 remaining ports won't take a month...
Before looking into this, I didn't know the CPU had that many available frequency multipliers, kinda wondering why it does. But hey, as an experiment, I think it at least has a little promise.