[CentOS] Extremely slow disk transfers with SATA disks

[Bob Chiodini]

On Thu, 2006-03-23 at 15:37 +0100, Vanja Hrustic wrote:
> I have installed CentOS 4.2, recently, on new computer.
> 
> Everything seemed to work fine.
> 
> However, we've found out that disk transfers are incredibly slow, and I
> just can't figure out what to do in order to fix it.
> 
> Computer is now running CentOS 4.3 (updated it 1 hour ago), and same
> thing is still present. I hoped kernel upgrade might fix it, but it
> didn't.
> 
> Computer has 2 SATA disks, 200GB (Maxtor 6Y200M0) and 80GB (Maxtor
> 6L080M0).
> 
> There are no errors or any other indications of problems in logs/dmesg.
> 
> Copying 500MB file takes around 210 seconds. Around 2.3MB/sec.
> 
> Doesn't matter if I am copying from one hard disk to another, or onto
> the same hard disk.
> 
> When copying, vmstat shows:
> 
>  r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
>  0  0    160 525928   1060 278044    0    0  1609  1112 1387  1146  9  4 39 47
>  0  0    160 525928   1060 278044    0    0     0     0 1084   745  0  0 100  0
>  0  0    160 525844   1060 278044    0    0     0     0 1067   795  2  0 98  0
>  0  0    160 525844   1060 278044    0    0     0     0 1084   745  1  0 99  0
>  0  0    160 525844   1072 278044    0    0     0   156 1073   757  0  0 100  0
>  0  1    160 525564   1148 278228    0    0   240     0 1129  1136  2  1 77 20
>  0  1    160 522708   1176 281256    0    0  1600     0 1283  1155  0  2  0 98
>  0  1    160 519380   1184 284568    0    0  1672     0 1327  1231  0  2  0 98
>  3  1    160 516404   1196 287368    0    0  1412     0 1264  1126  1  1  0 98
>  0  1    160 513524   1196 290424    0    0  1536     0 1353  1926 11  2  0 87
>  0  3    160 510132   1208 293544    0    0  1544  7348 1407  2163 24  6  0 71
>  0  1    160 513780   1216 289972    0    0  1412   140 1360  1536  9  3  0 88
>  2  0    160 515316   1220 288464    0    0  1104     0 1341  2067 47  4  0 49
>  0  1    160 518996   1216 284608    0    0  1712     0 1333  1676 34  7  0 59
>  0  1    160 516820   1224 286832    0    0  2052     0 1318  2025 25  4  0 71
>  1  2    160 512388   1228 291024    0    0  2952  8736 1433  1431  6  4  0 90
>  2  1    160 504772   1244 295304    0    0  3076    76 1382  1453 15  5  0 80
>  0  1    160 502340   1252 301096    0    0  3332     0 1358  1624  7  5  0 88
>  0  1    160 495236   1260 308216    0    0  3588     0 1323  1631  2  3  0 95
>  1  1    160 487620   1268 315640    0    0  3716     0 1355  1175  7  2  0 91
>  0  2    160 480708   1276 322120    0    0  3204 16732 1423  1354  2  5  0 93
> 
> You can see text version here (in case this got wrapped):
> 
> http://www.vanja.com/vmstat.txt
> 
> I've tried vmstat on my machine, with IDE disks, and I've noticed that I 
> ave much less 'cs' (context switches), although it might be because I'm
> not having as many processes in the background (problematic computer
> is running KDE :).
> 
> On my computer, number of blocks in (bi) gets into tens of thousands,
> while problematic computer has 1500-3000 bs/second.
> 
> Does anyone know what I might do in order to find what the problem
> might be?
> 
> I couldn't find any SATA specific tools, which might help with
> troubleshooting. hdparm -tT tests show:
> 
> /dev/sda:
>  Timing cached reads:   2408 MB in  2.00 seconds = 1202.98 MB/sec
>  Timing buffered disk reads:  150 MB in  3.02 seconds =  49.61 MB/sec
> 
> /dev/sdb:
>  Timing cached reads:   2628 MB in  2.00 seconds = 1313.54 MB/sec
>  Timing buffered disk reads:  166 MB in  3.02 seconds =  54.88 MB/sec
> 
> All filesystems are ext3.
> 
> This "problematic" computer has nforce3 motherboard, I have nforce2.
> Same CentOS, but no SATA on my computer.
> 
> This really affects complete performance of computer, and any help or
> ideas are welcome.
> 
> Btw, no problems with transfer under Windows :(
> 
> Thanks.
> 
> Vanja

Your hdparm numbers are slightly better than mine for the same drive.

For me:  copying a 3.2GB file achieved 18.4MB/s

I guess hdparm is not a real-world indicator.

What does top tell you?  Is something hogging the CPU.

Bob...