Boston Linux & Unix (BLU) Home | Calendar | Mail Lists | List Archives | Desktop SIG | Hardware Hacking SIG
Wiki | Flickr | PicasaWeb | Video | Maps & Directions | Installfests | Keysignings
Linux Cafe | Meeting Notes | Blog | Linux Links | Bling | About BLU

BLU Discuss list archive


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

Re: ATA over Ethernet (aoe) Experience?



 Peter Petrakis wrote: 
> Suppose that you don't have an investment in AoE equipment; why 
> would one chose AoE over iSCSI? 
>   

Correct, I don't have any aoe equipment, nor do I have any iscsi 
equipment.  And I don't imagine buying any of either.  But that doesn't 
mean I can't use the *protocols*... 

What I am investigating is how to do a "smallish amount of serving" with 
completely redundant equipment.  I do not like the idea of a single 
point of failure, even less do I like the idea of a specialized piece of 
hardware being a single point of failure.  This means a specialized rack 
of disks bugs me because the rack itself becomes a failure point--unless 
I can afford two of them.  In the "smallish amount of serving" 
department two complete disk racks is out of range. 

My current idea is two identical servers.  Each capable of carrying the 
entire load alone, so the other is the spare.  For $5K one can get a 
pretty decent machine (fast multicore CPU, lots of RAM, lots of disk). 
A second $5K buys the 100% redundant spare.  And everything can fit in 2U. 

So far, so good. 

The fuzzy part comes in when I figure out how to quickly switch between 
physical machines.  So maybe I make each machine an aoe target.  The 
result would be that if the running machine goes down the spare could be 
fired up, already having a complete copy of the data the dead machine 
was using. 

As for what I would be serving, there would be some server consolidation 
going on here, probably as virtual machines under Virtualbox. 

What I am still trying to understand is how, when switching a VM from 
one physical server to another, to not have a nasty delay as a 
multi-gigabyte disk image is copied over.  And, how to have resizable 
allocations between VMs.  And, how to have software Raid 1 to protect me 
from a disk dying.  And, have the ability to migrate one VM at a time, 
not just all-at-once.  Stacking all that up in the right order is a 
puzzle.  aoe might be part of the answer. 

As for trying aoe in person, it looks cool, but in my current 
distribution of choice (Ubuntu) aoe isn't playing nice with udev.  A 
worry is that there might not be a critical mass around aoe, it might 
become an orphan.  It seems to have been broken in Ubuntu for sometime 
and mostly no one noticed. 

As for iscsi, is it suitable for a computer-to-computer setup?  (An 
appeal of aoe is that it has a shorter stack: it can't be routed, it is 
simple looking, it can use efficient jumbo packets that will fit just 
fine on a crossover cable running between the two servers.) 


-kb 

-- 
This message has been scanned for viruses and 
dangerous content by MailScanner, and is 
believed to be clean. 

_______________________________________________ 
Discuss mailing list 
[hidden email] 
http://lists.blu.org/mailman/listinfo/discuss
 


BLU is a member of BostonUserGroups
BLU is a member of BostonUserGroups
We also thank MIT for the use of their facilities.

Valid HTML 4.01! Valid CSS!



Boston Linux & Unix / webmaster@blu.org