The Caffeinated Penguin

musings of a crackpot hacker

Catharsis

Posted By on July 7, 2007

(note, since case is the name of a machine (kudos if you get the reference), I will use “box” here where I mean “the piece of sheet metal into which computer components are put)

So, this evening I took a fairly radical step which I had been considering for some time – I gutted my main workstation (case) and put the boot drive and motherboard into the box containing the disk array and beefy PSU for the NFS server (yt). The idea here is that, as fast as gigabit ethernet is, having the disks be local will be faster.

The final straw came when, this week, two things happened: (1) The UPS to which case was attached dropped in a brownout, while I was working, rebooting case. yt's UPS stayed up. I suspect case's UPS has a bad battery. (2) One of yt's drives started to make evil clicking noises. Listening at the front of the box, it was on the side with only two drives – the boot drive and disk 8 (or 7 if you count from 0) on the RAID 5 array. Lovely.

Tonight I had some time, so I did the transplant. Since there wasn't going to be a gigE backbone anymore, those cards can come out. PSU can handle the additional load of the extra CPU and beefier video card without issue. The boot drive is going to be the drive which was in case, so the extra space will still be used for rsnapshot backups (in case I fat finger something).

In all, it took less time than I thought it would, though there are still some loose ends to tie up – namely cleaning up old vestiges of NFS/NIS infrastructure on case and replicating the services yt used to provide on case (SMB shares and NTP, primarily).

case's very nice box will not go to waste – it will be reused to make a new (as yet unnamed) machine based around a dual-core CPU of some variety, running modern SATA disks (I'm leaning towards a pair of 750's in a software mirror) using an Intel GMA X3000 integrated video (unless ATI or NVidia does something with F/OSS drivers before I buy) w/ 4GB of RAM and a beefy PSU which I already own.

I've also already cannibalized out the 2 512MB DIMMS which were in yt's motherboard and put them into a Mini-ITX board I had lying around (for safekeeping, you know). I'm debating what to do with this one – it's a perfectly serviceable machine, though it's weak in the graphics department. I'm debating if I should build it into something interesting (like the VW Beetle PC, only different), or otherwise put it to some type of practical use (X Terminal perhaps)?

The local disk access is much faster. Oh, and there is no more hard drive clicking.

(the difference in kernel versions gives some indication as to how long I’ve been considering this move). tests are done with tiobench --size 512 /home, NFS mount
Unit information
================
File size = megabytes
Blk Size  = bytes
Rate      = megabytes per second
CPU%      = percentage of CPU used during the test
Latency   = milliseconds
Lat%      = percent of requests that took longer than X seconds
CPU Eff   = Rate divided by CPU% - throughput per cpu load

Sequential Reads
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-15-generic             2048  4096    1   16.39 9.757%     0.237      254.79   0.00000  0.00000   168
2.6.20-15-generic             2048  4096    2   15.46 17.09%     0.503      564.42   0.00000  0.00000    90
2.6.20-15-generic             2048  4096    4   15.95 35.09%     0.970     1032.30   0.00000  0.00000    45
2.6.20-15-generic             2048  4096    8   15.83 70.40%     1.955     2096.14   0.00019  0.00000    22

Random Reads
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-15-generic             2048  4096    1    0.58 0.998%     6.704     1841.12   0.00000  0.00000    58
2.6.20-15-generic             2048  4096    2    1.12 3.418%     6.751       81.39   0.00000  0.00000    33
2.6.20-15-generic             2048  4096    4    1.90 13.34%     7.988       95.30   0.00000  0.00000    14
2.6.20-15-generic             2048  4096    8    2.22 32.84%    13.801      223.28   0.00000  0.00000     7

Sequential Writes
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-15-generic             2048  4096    1    6.05 7.012%     0.625     7319.08   0.00687  0.00000    86
2.6.20-15-generic             2048  4096    2    5.91 15.82%     1.291    11568.69   0.01030  0.00019    37
2.6.20-15-generic             2048  4096    4    5.85 35.79%     2.582    12656.48   0.02575  0.00134    16
2.6.20-15-generic             2048  4096    8    5.48 167.8%     5.032    17257.68   0.06981  0.00401     3

Random Writes
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-15-generic             2048  4096    1    0.41 0.418%     0.023        0.51   0.00000  0.00000    98
2.6.20-15-generic             2048  4096    2    0.40 0.873%     3.540    14020.55   0.02500  0.02500    45
2.6.20-15-generic             2048  4096    4    0.39 1.780%     8.481    10081.21   0.12500  0.02500    22
2.6.20-15-generic             2048  4096    8    0.35 3.307%    28.507     6889.94   0.60000  0.00000    11
/home, local disk
Unit information
================
File size = megabytes
Blk Size  = bytes
Rate      = megabytes per second
CPU%      = percentage of CPU used during the test
Latency   = milliseconds
Lat%      = percent of requests that took longer than X seconds
CPU Eff   = Rate divided by CPU% - throughput per cpu load

Sequential Reads
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-16-generic             512   4096    1  216.32 93.29%     0.016       38.59   0.00000  0.00000   232
2.6.20-16-generic             512   4096    2  372.12 344.5%     0.018       40.05   0.00000  0.00000   108
2.6.20-16-generic             512   4096    4  371.39 633.4%     0.033      116.09   0.00000  0.00000    59
2.6.20-16-generic             512   4096    8  401.29 1387.%     0.063      198.69   0.00000  0.00000    29

Random Reads
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-16-generic             512   4096    1  220.57 95.99%     0.015        1.42   0.00000  0.00000   230
2.6.20-16-generic             512   4096    2  412.60 380.2%     0.016        0.20   0.00000  0.00000   108
2.6.20-16-generic             512   4096    4  405.32 539.5%     0.025       18.27   0.00000  0.00000    75
2.6.20-16-generic             512   4096    8  350.26 358.6%     0.021       19.57   0.00000  0.00000    98

Sequential Writes
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-16-generic             512   4096    1   16.39 19.22%     0.096     2727.59   0.00076  0.00000    85
2.6.20-16-generic             512   4096    2    8.46 27.09%     0.361     3913.13   0.00763  0.00000    31
2.6.20-16-generic             512   4096    4    9.89 57.44%     0.601     7652.50   0.00763  0.00000    17
2.6.20-16-generic             512   4096    8   10.75 128.4%     0.956    11992.25   0.01221  0.00076     8

Random Writes
                              File  Blk   Num                   Avg      Maximum      Lat%     Lat%    CPU
Identifier                    Size  Size  Thr   Rate  (CPU%)  Latency    Latency      >2s      >10s    Eff
---------------------------- ------ ----- ---  ------ ------ --------- -----------  -------- -------- -----
2.6.20-16-generic             512   4096    1    0.86 0.726%     0.021        0.09   0.00000  0.00000   118
2.6.20-16-generic             512   4096    2    0.87 1.603%     0.027        0.57   0.00000  0.00000    54
2.6.20-16-generic             512   4096    4    0.60 2.385%     0.037       31.75   0.00000  0.00000    25
2.6.20-16-generic             512   4096    8    0.48 2.489%     0.024        0.22   0.00000  0.00000    19
This is a pretty huge throughput increase and decrease in latency.

Comments

Loading Facebook Comments ...

Leave a Reply

Please note: Comment moderation is currently enabled so there will be a delay between when you post your comment and when it shows up. Patience is a virtue; there is no need to re-submit your comment.