Hack and / - Linux Troubleshooting, Part I: High Load
The next cause for high load is a system that has run out of available RAM and has started to go into swap. Because swap space is usually on a hard drive that is much slower than RAM, when you use up available RAM and go into swap, each process slows down dramatically as the disk gets used. Usually this causes a downward spiral as processes that have been swapped run slower, take longer to respond and cause more processes to stack up until the system either runs out of RAM or slows down to an absolute crawl. What's tricky about swap issues is that because they hit the disk so hard, it's easy to misdiagnose them as I/O-bound load. After all, if your disk is being used as RAM, any processes that actually want to access files on the disk are going to have to wait in line. So, if I see high I/O wait in the CPU row in top, I check RAM next and rule it out before I troubleshoot any other I/O issues.
When I want to diagnose out of memory issues, the first place I look is the next couple of lines in the top output:
Mem: 1024176k total, 997408k used, 26768k free, 85520k buffers Swap: 1004052k total, 4360k used, 999692k free, 286040k cached
These lines tell you the total amount of RAM and swap along with how much is used and free; however, look carefully, as these numbers can be misleading. I've seen many new and even experienced administrators who would look at the above output and conclude the system was almost out of RAM because there was only 26768k free. Although that does show how much RAM is currently unused, it doesn't tell the full story.
When you access a file and the Linux kernel loads it into RAM, the kernel doesn't necessarily unload the file when you no longer need it. If there is enough free RAM available, the kernel tries to cache as many files as it can into RAM. That way, if you access the file a second time, the kernel can retrieve it from RAM instead of the disk and give much better performance. As a system stays running, you will find the free RAM actually will appear to get rather small. If a process needs more RAM though, the kernel simply uses some of its file cache. In fact, I see a lot of the overclocking crowd who want to improve performance and create a ramdisk to store their files. What they don't realize is that more often than not, if they just let the kernel do the work for them, they'd probably see much better results and make more efficient use of their RAM.
To get a more accurate amount of free RAM, you need to combine the values from the free column with the cached column. In my example, I would have 26768k + 286040k, or over 300Mb of free RAM. In this case, I could safely assume my system was not experiencing an out of RAM issue. Of course, even a system that has very little free RAM may not have gone into swap. That's why you also must check the Swap: line and see if a high proportion of your swap is being used.
If you do find you are low on free RAM, go back to the same process output from top, only this time, look in the %MEM column. By default, top will sort by the %CPU column, so simply type M and it will re-sort to show you which processes are using the highest percentage of RAM. In the output in Listing 3, I sorted the same processes by RAM, and you can see that the nagios2db_status process is using the most at 6.6%.
Listing 3. Processes Sorted by RAM
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 18749 nagios 16 0 140m 134m 1868 S 12 6.6 1345:01 nagios2db_status 9463 mysql 16 0 686m 111m 3328 S 53 5.5 569:17.64 mysqld 24636 nagios 17 0 34660 10m 712 S 8 0.5 1195:15 nagios 22442 nagios 24 0 6048 2024 1452 S 8 0.1 0:00.04 check_time.pl
I/O-bound load can be tricky to track down sometimes. As I mentioned earlier, if your system is swapping, it can make the load appear to be I/O-bound. Once you rule out swapping though, if you do have a high I/O wait, the next step is to attempt to track down which disk and partition is getting the bulk of the I/O traffic. To do this, you need a tool like iostat.
The iostat tool, like top, is a complicated and full-featured tool that could fill up its own article. Unlike top, although it should be available for your distribution, it may not be installed on your system by default, so you need to track down which package provides it. Under Red Hat and Debian-based systems, you can get it in the sysstat package. Once it's installed, simply run iostat with no arguments to get a good overall view of your disk I/O statistics:
Linux 2.6.24-19-server (hostname) 01/31/2009 avg-cpu: %user %nice %system %iowait %steal %idle 5.73 0.07 2.03 0.53 0.00 91.64 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 9.82 417.96 27.53 30227262 1990625 sda1 6.55 219.10 7.12 15845129 515216 sda2 0.04 0.74 3.31 53506 239328 sda3 3.24 198.12 17.09 14328323 1236081
Like with top, iostat gives you the CPU percentage output. Below that, it provides a breakdown of each drive and partition on your system and statistics for each:
tps: transactions per second.
Blk_read/s: blocks read per second.
Blk_wrtn/s: blocks written per second.
Blk_read: total blocks read.
Blk_wrtn: total blocks written.
By looking at these different values and comparing them to each other, ideally you will be able to find out first, which partition (or partitions) is getting the bulk of the I/O traffic, and second, whether the majority of that traffic is reads (Blk_read/s) or writes (Blk_wrtn/s). As I said, tracking down the cause of I/O issues can be tricky, but hopefully, those values will help you isolate what processes might be causing the load.
For instance, if you have an I/O-bound load and you suspect that your remote backup job might be the culprit, compare the read and write statistics. Because you know that a remote backup job is primarily going to read from your disk, if you see that the majority of the disk I/O is writes, you reasonably can assume it's not from the backup job. If, on the other hand, you do see a heavy amount of read I/O on a particular partition, you might run the lsof command and grep for that backup process and see whether it does in fact have some open file handles on that partition.
As you can see, tracking down I/O issues with iostat is not straightforward. Even with no arguments, it can take some time and experience to make sense of the output. That said, iostat does have a number of arguments you can use to get more information about different types of I/O, including modes to find details about NFS shares. Check out the man page for iostat if you want to know more.
Up until recently, tools like iostat were about the limit systems administrators had in their toolboxes for tracking down I/O issues, but due to recent developments in the kernel, it has become easier to find the causes of I/O on a per-process level. If you have a relatively new system, check out the iotop tool. Like with iostat, it may not be installed by default, but as the name implies, it essentially acts like top, only for disk I/O. In Listing 4, you can see that an rsync process on this machine is using the most I/O (in this case, read I/O).
Kyle Rankin is a director of engineering operations in the San Francisco Bay Area, the author of a number of books including DevOps Troubleshooting and The Official Ubuntu Server Book, and is a columnist for Linux Journal.
Free DevOps eBooks, Videos, and more!
Regardless of where you are in your DevOps process, Linux Journal can help!
We offer here the DEFINITIVE DevOps for Dummies, a mobile Application Development Primer, and advice & help from the expert sources like:
- Linux Journal
- Readers' Choice Awards 2014
- diff -u: What's New in Kernel Development
- How Can We Get Business to Care about Freedom, Openness and Interoperability?
- Cooking with Linux - Serious Cool, Sysadmin Style!
- Synchronize Your Life with ownCloud
- Days Between Dates?
- Computing without a Computer
- Non-Linux FOSS: Don't Type All Those Words!
- Autokey: Shorthand for Typists