[Aoetools-discuss] Strange performance issue with partitions

[Bernard L. <be...@va...>]

Hi all:

I have been experimenting with (k)vblade and AoE for Linux for the
last little while, and have encountered a strange performance issue.

On my "server", I created a 8GB file and set it up via /dev/loop1 and
I share it out with vblade as e1.0.

On my "client", I run fdisk and created 1 large extended partition and
5+ logical partitions, each around 1GB.

Again, on the client, I ran hdparm -tT on different partitions and am
getting the following odd behavior:

/dev/etherd/e1.0p11:
 Timing cached reads:   1880 MB in  2.00 seconds = 938.74 MB/sec
 Timing buffered disk reads:  200 MB in  3.00 seconds =  66.59 MB/sec

/dev/etherd/e1.0p12:
 Timing cached reads:   2904 MB in  2.05 seconds = 1418.88 MB/sec
 Timing buffered disk reads:  180 MB in  3.00 seconds =  59.95 MB/sec

/dev/etherd/e1.0p5:
 Timing cached reads:   2064 MB in  2.07 seconds = 995.33 MB/sec
 Timing buffered disk reads:  100 MB in  3.03 seconds =  33.01 MB/sec

Basically, partition 5 (first logical partition) performs poorly
(buffered disk reads), and all the rest (6-12) performs "optimally".

Does anybody know what is going on here?

Regarding the other discussion thread about needing to run multiple
vblade processes to share out separate EtherDrive blades, it is true
that with kvblade, one process could handle multiple blades, however,
in experience, I notice that the CPU load on the server running
(k)vblade is just as bad when there are a significant number of blades
being shared out.

I was wondering if there are plans to update the (k)vblade software
and perhaps to address some of the CPU consumption issues.

P.S. I am running vblade 14 and aoe6-46 with my tests, pretty much the
most up-to-date code.

P.P.S I have started a channel #aoe on irc.freenode.net hoping to
attract some discussions on AoE - so far there is one other user who
drops by reguarly, hope to see you folks there!

Thanks!

Bernard