Slow SCSI perf in RH 7.3
I have a server:
2.4.20-28.7 #1 Thu Dec 18 11:31:59 EST 2003 i686
with SCSI hard disks (not raid):
SCSI subsystem driver Revision: 1.00
scsi0 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.2.8
<Adaptec 3960D Ultra160 SCSI adapter>
aic7899: Ultra160 Wide Channel A, SCSI Id=7, 32/253 SCBs
scsi1 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.2.8
<Adaptec 3960D Ultra160 SCSI adapter>
aic7899: Ultra160 Wide Channel B, SCSI Id=7, 32/253 SCBs
blk: queue dfdbe014, I/O limit 4095Mb (mask 0xffffffff)
Vendor: FUJITSU Model: MAP3367NP Rev: 5605
Type: Direct-Access ANSI SCSI revision: 03
blk: queue dfdbe214, I/O limit 4095Mb (mask 0xffffffff)
Vendor: FUJITSU Model: MAP3367NP Rev: 5605
Type: Direct-Access ANSI SCSI revision: 03
blk: queue dfdbe614, I/O limit 4095Mb (mask 0xffffffff)
Vendor: SEAGATE Model: ST373307LW Rev: 0005
Type: Direct-Access ANSI SCSI revision: 03
blk: queue dfdbea14, I/O limit 4095Mb (mask 0xffffffff)
scsi0:A:0:0: Tagged Queuing enabled. Depth 32 scsi0:A:1:0: Tagged Queuing enabled. Depth 32 scsi0:A:4:0: Tagged Queuing enabled. Depth 32
Attached scsi disk sda at scsi0, channel 0, id 0, lun 0 Attached scsi disk sdb at scsi0, channel 0, id 1, lun 0 Attached scsi disk sdc at scsi0, channel 0, id 4, lun 0 (scsi0:A:0): 160.000MB/s transfers (80.000MHz DT, offset 127, 16bit) SCSI device sda: 71132959 512-byte hdwr sectors (36420 MB) Partition check:
sda: sda1 sda2 sda3 sda4 < sda5 sda6 sda7 sda8 sda9 >
(scsi0:A:1): 160.000MB/s transfers (80.000MHz DT, offset 127, 16bit) SCSI device sdb: 71132959 512-byte hdwr sectors (36420 MB) sdb: sdb1 (scsi0:A:4): 160.000MB/s transfers (80.000MHz DT, offset 63, 16bit) SCSI device sdc: 143374744 512-byte hdwr sectors (73408 MB) sdc: sdc1 Journalled Block Device driver loaded
used for graphic files server to Apples (via Netatalk 1.5.2-3) with ext3. OS on one disk, data files on the other two. IDE removable storage.
It's great with only one user, but with multiple Apple users, it does a passable impression of a boat anchor. Even ssh then comes back one character at a time. Top shows nothing out the ordinary, I think. Here's some typical 'sar -b' stuff:
03:20:00 PM tps rtps wtps bread/s bwrtn/s 03:30:00 PM 3.97 1.97 2.00 75.65 37.09 03:40:00 PM 4.41 2.32 2.09 336.19 122.32 03:50:00 PM 2.93 1.87 1.05 326.00 131.52 04:00:00 PM 2.44 1.80 0.65 294.09 27.97 04:10:00 PM 2.16 0.58 1.57 107.37 53.69 04:20:00 PM 0.81 0.16 0.66 32.26 31.38 04:30:00 PM 1.78 0.72 1.06 84.34 94.09 04:40:00 PM 1.28 0.55 0.73 55.81 97.88 04:50:00 PM 0.69 0.21 0.48 3.78 13.29 05:00:01 PM 1.82 0.51 1.32 6.02 51.81 05:10:00 PM 1.22 0.04 1.18 0.72 89.07 05:20:00 PM 2.60 1.13 1.47 63.19 30.13 05:30:00 PM 1.36 0.31 1.05 3.44 46.59 Average: 2.51 1.57 0.94 42.79 32.36
which I wish I knew well enough to interpret. :-(
So, I'd like to do *something* to speed things up. The likely suspects are:
1) Tweak the kernel somehow, in the knowledge that I'll probably just make things worse (caching?)
2) Buy RAID (which level do I want? Is SATA raid faster than SCSI non-raid?)
3) Buy nicer/more expensive hard disks on the current controller, or, perhaps, a faster non-raid controller.
All the connections, terminators, etc. seem happy.
I would consider trading a little reliability for a lot of performance.
The inevitable omitted information upon request. I was hoping for some suggestions from some very kind list members.
Thanks!
Signature · 5 lines
_________________________________________ Nachman Yaakov Ziskind, FSPA, LLM [email protected] Attorney and Counselor-at-Law http://ziskind.us Economic Group Pension Services http://egps.com Actuaries and Employee Benefit Consultants
A message that is not in the archive (sent off-list or not yet received): Pine.LNX.4.62.0503301758400.1159@morpheus
Burton Windle wrote (on Wed, Mar 30, 2005 at 06:01:21PM -0500):
Quoted 10 lines · Nachman Yaakov Ziskind
Quoted 4 lines
I have a server:
2.4.20-28.7 #1 Thu Dec 18 11:31:59 EST 2003 i686Looking at the output of 'top' may be helpful, as it will show if the system is CPU or IO bound. However, plese note that 2.4.20 is quite old (November of 2002), and contains numerous security holes. Upgrading to something more recent, like 2.4.29, may be a good first start.
Thanks for the quick reply. Sar seems to indicate that the server is disk-bound (although some swapping is going on, so more RAM is on the way). CPU is idle, consistently, 99% of the time.
Would upgrading the kernel help the disk performance? Server is not connected to the internet, so security concerns are not *that* important, but performance is ...
Signature · 5 lines
_________________________________________ Nachman Yaakov Ziskind, FSPA, LLM [email protected] Attorney and Counselor-at-Law http://ziskind.us Economic Group Pension Services http://egps.com Actuaries and Employee Benefit Consultants