1 |
Hi Brian, |
2 |
|
3 |
Brian Kroth wrote: |
4 |
> I currently manage about 40 Window, OSX, and Hardened Gentoo servers. |
5 |
> I will soon have 12 P4 servers that were previously used as video |
6 |
> encoders free as well as an Apple XRaid. With all this spare hardware |
7 |
> I thought I'd research setting up a cluster of servers running Apache |
8 |
> for load balancing and high availability. I'm also looking into a |
9 |
> MySQL cluster, but that wouldn't require a shared filesystem. I'm |
10 |
> wondering if anyone has done something like this before and in |
11 |
> particular knows a good filesystem to use so that each of the servers |
12 |
> can access and potentially write to the same storage array. I've |
13 |
> accomplished the same thing with XServes running OSX, but they like to |
14 |
> charge you a pretty penny for the XSan software that allows this which |
15 |
> I thought I'd try to avoid if possible. So far I've seen only GFS, |
16 |
> but haven't gotten much reading done on it yet. Any other tips or |
17 |
> insights would be appreciated as well. |
18 |
|
19 |
I currently manage 4 different shared storage systems on gentoo with |
20 |
SAN-storage based on AoE (coraid 1520 machines) |
21 |
We choose GFS for the filesystem. |
22 |
|
23 |
It works, although the throughput is not spectacular. |
24 |
This has partly to do with GFS and partly because the underlying storage |
25 |
hardware doesn't support faster throughput. |
26 |
If you have any specific questions about GFS I'd be happy to answer them. |
27 |
|
28 |
Apart from that we're currently planning a Lustre deploy for fast-storage. |
29 |
This is still very much in alpha, so no comments yet. |
30 |
|
31 |
I've looked at implementing a mysql-cluster in great detail and let it |
32 |
go because it's mostly meant for a setup with a fixed database-size. |
33 |
If you experience (large) growth in your database volume clustering is |
34 |
not very suitable (we're in that situation) |
35 |
We're currently load-balancing our databases with LVS (ipvsadm) and a |
36 |
multi-master replication setup. |
37 |
This works out really well for us, serving estimated average 15Kq/s |
38 |
with 50 servers on a very large dataset (> 40Gb) and a fairly random |
39 |
read pattern. |
40 |
|
41 |
I would strongly disadvise running multiple databases off GFS. |
42 |
>From what I gathered it works well for active/passive failover sets |
43 |
and/or active/active failover, but performance will be hurt compared to |
44 |
a "normal" fs. |
45 |
|
46 |
Regards, |
47 |
|
48 |
Ramon |
49 |
|
50 |
-- |
51 |
gentoo-server@g.o mailing list |