[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: i-ram vs. tmpfs (was: Re: Mail clustering)

> > > On Wed, Apr 04, 2007 at 11:50:00PM +0000, Andy Smith wrote:
> > > > If you have 4G of RAM to put in an i-RAM then why would you use it
> > > > as swap?  If your machine swaps, use the 4G as real RAM!
> > 
> > On 05.04.07 10:05, Craig Sanders wrote:
> > > exactly.

> On Thu, Apr 05, 2007 at 12:30:10PM +0200, Matus UHLAR - fantomas wrote:
> > Are you both sure that the speed of accessing the data thgourh SATA
> > interface won't slow down your system more than using it as swap device
> > in addition to journals?

On 05.04.07 22:19, Craig Sanders wrote:
> we're both saying NOT to use it as a swap device, but to put that 4GB
> RAM into the system instead (or as well as, if you need both).

and I'm asking if you are sure that will not slow down your system speed.
The access to this memory will be slower than access to your main RAM,
because it's limited with PCI/SATA bus speed. Using that memory as RAM may
slow down your system.

> but swapping to a ram-disk is just silly unless your M/B can't take any
> more ram.

I remember there were problems with some chipsets only able to cache limited
amount of system RAM. Using anything above that RAM did slow down the system
speed. That's why linux has CONFIG_MTD_SLRAM option, which can at least use
that memory for storage or swap (read kernel docs for more info)

So I'm asking you again if you are sure that using i-RAM as direct memory
will not slow down your system as described above.

> > > i.e. spend $50 on a new SATA drive for the external journal device
> > > and get 80-90% of the performance improvement that you would get for
> > > spending $500 on a 4GB I-RAM.
> > 
> > the problem is probably the SATA bus speed... 
> no, it's not.  it's diminishing returns.
> moving the journal onto another device - ANY other device - gives you
> a big benefit. once you've done that, even moving it to a much faster
> device doesn't give you anywhere near the same additional benefit.

depends on amount of data journalled and speed of the device. Try moving
journal to floppy if you disagree :)

> > but even you say it's faster :)
> yes, using an SSD for fs journalling *WILL* be faster than using a
> hard-disk. it will also cost a *LOT* more than a hard disk, 10 times
> more using my figures above (and a lot more than that if you use a
> "real" SSD rather than an I-RAM).

I agree, I didn't oppose to this argument.

> for most people, the small extra additional benefit isn't worth the much
> greater cost.
> depending, of course, on exactly how much you need a few extra %
> performance, and how much money you can afford to throw at it.

I was talking about reliability too, which is much better, since the writes
are much faster, and since you can journal data that way (and the latter
will make the speed impact more noticeable)[A
Matus UHLAR - fantomas, uhlar@fantomas.sk ; http://www.fantomas.sk/
Warning: I wish NOT to receive e-mail advertising to this address.
Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu.
The early bird may get the worm, but the second mouse gets the cheese. 

Reply to: