Results 1 to 9 of 9
Thread: How high is a high iowait?
-
03-09-2011, 11:35 AM #1WHT Addict
- Join Date
- May 2008
- Posts
- 105
How high is a high iowait?
what percentage of iowait is considered to be high?
-
03-09-2011, 12:39 PM #2Web Hosting Evangelist
- Join Date
- Apr 2009
- Location
- Romania
- Posts
- 473
This could help:
http://serverfault.com/questions/126...what-iowait-isPidginHost.com - Managed dedicated servers, cloud servers and software development.
-
03-09-2011, 01:29 PM #3WHT Addict
- Join Date
- Aug 2008
- Posts
- 133
The answer is it depends. On a system like a production database, any persistent IO wait would indicate a likely disk bottleneck that should be immediately addressed. On a front-end webserver type box, perhaps 2-3% would be okay. On a storage backed, where bulk data is read/written, we can see IO wait percentages up to about 20%.
We use ganglia: http://ganglia.sourceforge.net/ to track and graph this and other metrics over time.
-
03-09-2011, 05:56 PM #4Web Hosting Master
- Join Date
- Mar 2008
- Location
- Los Angeles, CA
- Posts
- 555
Generally better to go off %util rather than iowait. Use iostat -xm 5 (for 5 sec averages) or a higher number for higher averages.
High iowait depends on how many cores your CPU has. 50% iowait on a dual core is equivelent to 12.5% iowait on a 8 core system. If you have a raid array it can sometimes be a bit deceiving because you could be at 100% util or 25% iowait on a quad core machine but only because you are maxing it at queue depth of 1 but use a multi-threaded I/O application (linux asychronous I/O) and you can actually push way more I/O when you have multiple requests happening at once.
If you are not sure yourself you can post some iostat -xm output and your servers config and I am sure some people here can give you an idea of if its being I/O maxed or heavily loaded on I/O.
-
03-09-2011, 07:30 PM #5WHT Addict
- Join Date
- May 2008
- Posts
- 105
This is from my dedi db server
avg-cpu: %user %nice %system %iowait %steal %idle
1.88 4.01 2.35 2.53 0.00 89.24
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 14.71 290.21 247.37 34397227 29318984
sda1 0.00 0.02 0.00 2234 56
sda2 0.24 0.09 46.63 10429 5526800
sda3 11.93 271.68 150.53 32200069 17841912
sda4 0.00 0.00 0.00 8 0
sda5 1.40 14.55 30.72 1724202 3640728
sda6 0.00 0.01 0.00 1243 272
sda7 0.98 3.02 18.71 357786 2217064
sda8 0.00 0.01 0.00 816 0
sda9 0.15 0.84 0.78 99832 92152
-
09-19-2011, 10:22 AM #6Aspiring Evangelist
- Join Date
- Mar 2009
- Posts
- 391
Would this be considered a high iowait ?
Linux 2.6.18-194.3.1.el5 (CentOS-55-64-minimal) 09/19/2011
avg-cpu: %user %nice %system %iowait %steal %idle
0.43 0.00 18.58 5.00 0.00 75.98
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 63.29 2242.76 2467.82 1995244939 2195473600
sda1 0.00 0.00 0.00 1357 328
sda2 0.00 0.00 0.00 988 48
sda3 63.28 2242.75 2467.82 1995241874 2195473224
sdb 47.97 1495.48 1850.54 1330437334 1646311296
sdb1 47.97 1495.48 1850.54 1330436430 1646311296
-
09-21-2011, 01:20 AM #7
I would use "iostat -x 10 10"
The first output is the average since the system booted, and after 10 seconds it will show you the current load. The far right column, util%, will tell you the percentage of time the disk is actively servicing requests. Under 40% will normally give you fine performance. Sometimes you can get away with higher, and sometimes even at 40% it will be a bit slow, but it's my rule of thumb.
The other column to look at is await. This is the average time a disk request will wait to be serviced. If you're seeing high values here like 200ms, then even if util% is low you may see unnecessary latency in your applications. This was an issue for one of our customers doing minecraft which does a lot of disk writing, that their await was maybe 300ms which caused a lot of lag. Even though their util% was low, the performance was poor until we upgraded them to SSD. But in most cases you can just look at util%IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers in under an hour
★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
Contact Us: sales@ioflood.com ★
-
09-21-2011, 04:56 AM #8Aspiring Evangelist
- Join Date
- Mar 2009
- Posts
- 391
Thanks for your valuable input funkywizard, this post has over 90,000 views, this is definitely a topic that needs attention.. I would love to see managed hosts post their opinion on high iowait ...
-
09-21-2011, 10:42 AM #9Web Hosting Master
- Join Date
- Nov 2009
- Location
- Cincinnati
- Posts
- 1,585
Depends on what cpu scheduled your using and what the servers doing as well.
In a heavy multi thread environment like openvz/xen an iowait of 10-15% can be terribly slow.
I try to keep most systems around 3-6% iowait average. Any more and I shove more disk into it.Last edited by Visbits; 09-21-2011 at 10:55 AM.
'Ripcord'ing is the only way!
Similar Threads
-
CentOS 5.3,vsftpd, high iowait, high LOAD :(
By ignitionservers in forum Hosting Security and TechnologyReplies: 25Last Post: 09-28-2009, 06:55 PM -
High Load from high iowait
By placebo in forum Hosting Security and TechnologyReplies: 10Last Post: 07-02-2008, 06:36 AM -
high load/high iowait/ once a day
By dsotmoon in forum Hosting Security and TechnologyReplies: 13Last Post: 04-11-2006, 06:39 PM -
High iowait
By Patryk in forum Dedicated ServerReplies: 2Last Post: 11-29-2005, 01:51 AM -
High iowait
By techwide in forum Hosting Security and TechnologyReplies: 5Last Post: 06-10-2005, 09:03 AM