A few observations on this:

On 6/23/2012 12:47 AM, Sam Varshavchik wrote:
> ...
>
> Now, for this discussion, I've been looking at the solid state disk 
> industry, and its slow, but steady, progress. With the prevalence of 
> SSDs, there's going to come a point where the platter speed and 
> average seek times become less, and less important.

I think it will still be a few years before they reach the $/GB mark of 
magnetic storage, but it might be just 2 to 5 years into the future.

>
> When that happens, the winner's going to be whoever gets out of the 
> way faster, and doesn't diddle around too long, before sending your 
> bytes to the disk, and doesn't waste its time getting them back to 
> you. Because, diddling around, and trying to optimize your file 
> layout, and making sure stuff isn't fragmented on disk; all of that, 
> is not going to be as important as it's used to be, that's where I 
> think we'll end up soon.

While the SSD industry is currently using "translation layers" that 
diddle around to hide the fundamental principles of SSD by emulating the 
interfaces of rotating platters, best performance from SSDs would need 
to work on the real world properties of such, and possibly some "inverse 
diddling" to compensate for inefficiencies in the "translation layers" 
on devices that cannot turn them off.

Linux JFFS2 was a good early solution but may be optimized/limited to 
smaller capacities.  Linux ext4 over translation layers may be the 
currently most common "cop out" solution, but consists of 3 layers, each 
optimized for a different situation.

>
> And I have to think that there's going to be a point at which the same 
> thing is going to apply at the higher levels in the stack. Meaning 
> that, if you're keeping index files around some sorts, and, trying to 
> optimize your calls into the filesystem, then there has to be a point 
> at which you're spending more time doing that, than what you actually 
> end up saving.

In general yes, but one factor that remains is the need to compensate 
for lower layers that refuse to get out of the way, such as slow file 
systems, bus/network contention etc.  Basic caching of stuff that 
happens to be slow to access on a given system (configurable or detected 
by measurement) is still a good technique, as long as you keep the cache 
much smaller than the full data set.

Here the idea (allegedly in Dovecot) of caching the message and UID 
lists locally on each node rather than reconstructing it from the 
off-node maildir on each request seems a sane solution, regardless of 
file storage technology.

>
> But, that's just where I think the wind is blowing. I had no way of 
> knowing the emergence of SSDss, of course, a decade ago. It's a 
> relatively recent phenomenon.
>

-- 
Jakob Bohm, CIO, partner, WiseMo A/S. http://www.wisemo.com
Transformervej 29, 2730 Herlev, Denmark. direct: +45 31 13 16 10 
<call:+4531131610>
This message is only for its intended recipient, delete if misaddressed.
WiseMo - Remote Service Management for PCs, Phones and Embedded

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Courier-imap mailing list
[email protected]
Unsubscribe: https://lists.sourceforge.net/lists/listinfo/courier-imap

Reply via email to