On 12/29/2011 01:39 PM, Joseph P Villa wrote: > Hi Dimitri and W, > > Here is my IOSTAT output.. > > Device: rMB/s wMB/s avgrq-sz avgqu-sz await svctm %util > sda 0.16 0.11 324.94 0.12 72.83 3.69 0.62 > sdb 0.09 0.01 109.57 0.02 13.18 3.41 0.62 > dm-0 0.00 0.00 8.00 0.00 91.41 14.42 0.01 > dm-1 0.00 0.00 9.80 0.00 28.22 7.95 0.11 > dm-2 0.00 0.00 8.36 0.04 78.18 4.57 0.25 > dm-3 0.09 0.01 85.74 0.09 41.36 1.49 0.34 > dm-4 0.00 0.00 14.90 0.01 71.61 2.61 0.02 > dm-5 0.16 0.11 18.37 0.18 6.06 0.21 0.62
Well, I've seen worse... though I'm not sure what the dm's are and where's the 3rd physical drive for your raid-5. Just for comparison, here's numbers from an nfs file server w/ raid-10 over "desktop"-level sata drives: Device: rMB/s wMB/s avgrq-sz avgqu-sz await svctm %util sda1 0.23 0.08 51.51 0.07 5.85 3.93 4.84 sdb1 0.25 0.08 29.21 0.11 4.53 3.25 7.58 sdc1 0.24 0.07 52.88 0.07 5.46 3.70 4.52 sdd1 0.22 0.09 33.58 0.09 4.87 2.84 5.35 sde1 0.26 0.09 23.35 0.10 3.15 2.23 6.75 sdf1 0.24 0.07 28.15 0.11 4.74 3.32 7.60 md0 1.44 0.23 20.29 0.00 0.00 0.00 0.00 drbd0 1.44 0.23 20.84 5.18 31.68 1.56 25.73 > I'm using fastcgi right now. I do get an error about wrong python version > in my httpd logs, could that have anything to do with it? ... > expected '2.6.5', found '2.6.6'. I doubt it. I think mod_wsgi can be faster esp. if you fine-tune it to your load, but I've no numbers to back that up. > We have between 10 and 20 users between instances on each Trac instance. > Would moving to postgres provide a huge benefit? Between transaction control, parallel processing, and caching I'd expect a noticeable speed-up with postgres. And put the db on some device that can do 100KB/s w/ < 50ms wait time. -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu
signature.asc
Description: OpenPGP digital signature
