My experience:<br>2 perlbal -> 2 trackers -> 2 apache/mod_php -> 2 mysql (1 master, 1 replication slave) <br>2 storage node ( 8 sata 300g disks total)<br>now store 2139276 files (include html, jpg ,xml...all files in my blog service)
<br><br>Checking devices...<br> host device size(G) used(G) free(G) use%<br> ---- --------------- ---------- ---------- ---------- ------<br> [ 1] dev1 267.556 25.126
242.430 9.39%<br> [ 1] dev2 275.073 25.821 249.252 9.39%<br> [ 1] dev3 275.073 25.467 249.606 9.26%<br> [ 1] dev4 275.073 25.885 249.188
9.41%<br> [ 2] dev11 264.183 4.712 259.471 1.78%<br> [ 2] dev12 279.450 4.941 274.509 1.77%<br> [ 2] dev7 275.073 39.244 235.829 14.27%
<br> [ 2] dev8 275.073 39.258 235.816 14.27%<br> ---- --------------- ---------- ---------- ---------- ------<br> total: 2186.555 190.453 1996.102 8.71%<br>Works fine, and you can find that dev11 dev12 is replaced 2 disks failed last month.
<br>Since min devcount=2 so, we didn't lose any files.<br>And the dev weight setting , let us balanced the files write to all disks.<br>But I've found the write speed is not fast enough for my service.<br>So I'm redesigning my storage node.
<br>Must lower power, lower price, and get more and more spindles as we can.<br>The reference is Petabox using in <a href="http://www.archive.org">http://www.archive.org</a> (<a href="http://www.archive.org/web/petabox.php">
http://www.archive.org/web/petabox.php</a>).<br>I think the mysql will not the bottlenack, but single point failure maybe the problem.<br>I've not found good opensource solution to do auto fail-over with mysql node.<br><br>
And I must thank livejournal provide us a good storage solution!<br><br><br>