1 |
there are many way to do clustering and one thing that i would consider |
2 |
a "holy grail" would be something like pvm [1] |
3 |
because nothing else seems to have similar horizontal scaling of cpu at |
4 |
the kernel level |
5 |
|
6 |
i would love to know the mechanism behind dell's equallogic san as it |
7 |
really is clustered lvm on steroids. |
8 |
GFS / orangefs / ocfs are not the easiest things to setup (ocfs is) and |
9 |
i've not found performance to be so great for writes. |
10 |
DRBD is only 2 devices as far as i understand, so not really super scalable |
11 |
i'm still not convinced over the likes of hadoop for storage, maybe i |
12 |
just don't have the scale to "get" it? |
13 |
|
14 |
the thing with clusters is that you want to be able to spin an extra |
15 |
node up and join it to the group and then you increase cpu / storage by |
16 |
n+1 but also you want to be able to spin nodes down dynamically and go |
17 |
down by n-1. i guess this is where hadoop is of benefit because that is |
18 |
not a happy thing for a typical file system. |
19 |
|
20 |
network load balancing is super easy, all info required is in each |
21 |
packet -- application load balancing requires more thought. |
22 |
this is where the likes of memcached can help but also why a good design |
23 |
of the cluster is better. localised data and tiered access etc... kind |
24 |
of why i would like to see a pvm kind of solution -- so that a page |
25 |
fault is triggered like swap memory which then fetches the relevant |
26 |
memory from the network: bearing in mind that a computer can typically |
27 |
trigger thousands of page faults a second and that memory access is very |
28 |
very many times faster than gigabit networking! |
29 |
|
30 |
[1] http://www.csm.ornl.gov/pvm/pvm_home.html |