Ubuntu 14.04 LTS supports TRIM on SSD drives

msataWe have been writing about “5 crucial optimizations for SSD usage in Ubuntu Linux” in the past. Today everybody is using SSD’s and the usage of the Linux “fstrim” command and the fstab “discard” options are well-known. Still it is interesting when the real guru’s talk about this technology. Martin Pitt and Colin King are such real gurus, they work with Canonical on very cool and very techy Linux things ūüėČ So when they speak about SSD support in Ubuntu, we are certainly listening!

TRIM support in Ubuntu 14.04 LTS

In the past 4 days there has been an interesting discussion on the SSD support in Ubuntu. It started with Martin Pitt who anounced that Ubuntu 14.04 LTS will have out of the box support for TRIM commands on SSD drives. Read the interesting parts of the discussion here:

Martin Pitt: “Finally, SSDs are now being trimmed automatically out of the box. Embarrassingly late, but at least in time for 14.04 LTS.”

Robert Bruce Park: “So, should I take ‘discard’ out of my fstab then?”

Martin Pitt: “You can keep it if you prefer (the cron job will ignore mounts with “discard”), but on a desktop it seems generally preferrable to run from cron.”

David James:¬† “How much of a performance/battery life penalty is there running discard?”

Colin King: “It’s hard to say as it depends on the SSD and firmware, but we did see mass file deletions (such as removing entire kernel source trees) taking significantly longer with discard on some devices that don’t support queued TRIM.¬†¬† Also, some actions like heavy random file extend/shrink with ftruncate() were 3.5-4 x slower with discard on ext4.¬†¬† Generally speaking though, discard overhead was minimal, but it does seem more efficient to batch up discards with periodic use of fstrim .”

Derek Dickerson: “I thought TRIM support with modern controllers causes ext IO\premature ware and tear and that garbage collection was something done on chip for modern SSD’s? Can someone tell me why everyone thinks the kernel should be worrying about this what the drives do by themselves? ”

Martin Pitt: “@Derek Dickerson , the drive has no idea about which blocks the OS considers as unused. Disk drives don’t “think” in terms of file systems, just in terms of reading and writing blocks.”

Matthew Eaton: “My only concern with this is fstrim will cause a good 1-2 minutes of heavy I/O activity causing the system to slow down while the user has no idea what is going on or why.”

Source: Martin Pitts Google+ page

About “swappiness” and “noatime”

It is good that TRIM will be enabled by default in the future. But this does not mean that Ubuntu will¬†configure your SSD as it should. It is only one of the “5 crucial optimizations for SSD usage in Ubuntu Linux“. Other things – like “swappiness” and “noatime” – are also important to get the best performance and the longest endurance out of your solid state disk.

Share

5 crucial optimizations for SSD usage in Ubuntu Linux

msataI bought my first SSD more than 5 years ago (late 2007), for my white¬†MacBook Core2Duo 2.0 Ghz. It may be needless to say that my MacBook ran faster than many MacBook Pro’s at that time. The look on peoples faces when they saw my MacBook boot faster than their MacBook Pro was unforgettable.

By that time I was also experimenting with running 2 x 8GB CompactFlash with SATA converters in RAID-0 mode in my completely silent (but old and under-clocked) Pentium 3 Linux desktop. When people would visit my home they couldn’t believe my PC was actually turned on: it was dead silent.

Even today I still enjoy building low-power, silent and ultra-fast (Linux) computers with an SSD as the primary drive. Nowadays the new mSATA standard (see picture) is being used. During the past years I learned a thing or two about configuring Linux on a SSD drive. Today I will share those things with you:

1) Buy more than enough RAM

Yep, you should have at least 16GB of RAM. I know you’ll say that you don’t use that, but that is exactly the point. If you would be using it, it would not be enough. And why not? RAM is relatively cheap these days and you don’t want your system to start swapping, since this will wear out your expensive SSD.

2) Use the default partition layout

Contrary to popular belief, the default (guided) partitioning will do. If you partition manually, keep in mind that you do need a swap partition that is bigger than your RAM size.

3) Avoid using swap

You have plenty of RAM, so you could turn off swap or even create a system without swap. I have been running without swap for a year or so and never experienced significant problems (other people say the same). The most annoying side-effect is that the hibernate function is not available, since it uses the swap partition. This is why I recommend to not completely disable swap, but to reduce the swappiness of Linux. This way Linux will use all available RAM before starting to write to your precious SSD, while still allowing you to hibernate. Still, be aware when it comes to using hibernate, because it uses a lot of writes. Run this one-liner to add a permanent swappiness setting:

echo -e "vm.swappiness=0" | sudo tee -a /etc/sysctl.conf

This will only be effective after reboot.

4) Disable access time logging

Remove the access time logging on the file system. Without the “noatime” flag on your file system every read will cause a write, because the file system will update the access time. This is bad for the life-time of your SSD, since it supports a limited number of writes and this is causing significantly more writes. Edit the “fstab” to add the “noatime” flag:

sudo nano /etc/fstab

Now change “errors=remount-ro” to “noatime,errors=remount-ro”. Save the file and reboot.

5) Enable TRIM

TRIM will stop your SSD from slowing down after using it for a while. If you do not enable TRIM the writes may become slower due to the erasing of the deleted blocks. By running this one-liner you can add TRIM as a daily cron job:

echo -e "#\x21/bin/sh\\nfstrim -v /" | sudo tee /etc/cron.daily/trim

You have to make the cron job executable using:

sudo chmod +x /etc/cron.daily/trim

It will be run every day to avoid slowing down your writes and you will hardly notice it.

More advice: monitor the expected life-time

Your disk will not run forever. It has an expected life-time and you should probably monitor it using:

sudo smartctl -data -A /dev/sda

This will output something like this:

maurits@nuc:~$ sudo smartctl -data -A /dev/sda
smartctl 5.41 2011-06-09 r3365 [x86_64-linux-3.8.0-26-generic] (local build)
Copyright (C) 2002-11 by Bruce Allen, http://smartmontools.sourceforge.net

=== START OF READ SMART DATA SECTION ===
SMART Attributes Data Structure revision number: 18
Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
1   Raw_Read_Error_Rate     0x0000   006   000   000    Old_age   Offline      -       6
3   Spin_Up_Time            0x0000   100   100   000    Old_age   Offline      -       0
4   Start_Stop_Count        0x0000   100   100   000    Old_age   Offline      -       0
5   Reallocated_Sector_Ct   0x0000   100   100   000    Old_age   Offline      -       0
9   Power_On_Hours          0x0000   100   100   000    Old_age   Offline      -       2592
12  Power_Cycle_Count       0x0000   100   100   000    Old_age   Offline      -       258
232 Available_Reservd_Space 0x0000   100   100   000    Old_age   Offline      -       4914564640
233 Media_Wearout_Indicator 0x0000   100   000   000    Old_age   Offline      -       100

maurits@nuc:~$

At ID# 233 you see the MWI. This is a value starting at 100 and when it reaches values below 10 you should start to worry. Note that different manufacturers may have different names and numbers for this indicator.

Final advice: avoid buying TLC drives

Triple Level Cell (TLC) drives are making their entry in the market. These drives have a dramatically lower expected life-time. The life-time of a drive is measured in Program/Erase (PE) cycles or in Peta Bytes Written (PBW). Check the specs carefully to avoid buying a cheap drive, thinking you made a good deal, while actually buying a model with a lower expected life-time.


Share