[Beowulf] SSD performance
Jonathan Engwall
engwalljonathanthereal at gmail.com
Sun Jul 22 11:09:23 PDT 2018
Ariel,
Sorry I cannot share any images. It was crashing through the night. Before
I write the new boot I will take a screenshot if possible.
On Sun, Jul 22, 2018, 6:52 AM ariel sabiguero yawelak <asabigue at fing.edu.uy>
wrote:
> Cheap drives (I use Crucial, AMD Radeon and Kingston) accept 3.000
> rewrites per memory block. There are a few of interesting attributes you
> can check from SMART. These are from my laptop (Crucial_CT960M500SSD1)
>
>
> ID# ATTRIBUTE_NAME FLAG VALUE WORST THRESH TYPE UPDATED
> WHEN_FAILED RAW_VALUE
> 9 Power_On_Hours 0x0032 100 100 000 Old_age
> Always - 8611
> 12 Power_Cycle_Count 0x0032 100 100 000 Old_age
> Always - 4800
> 173 Ave_Block-Erase_Count 0x0032 096 096 000 Old_age
> Always - 134
> 180 Unused_Reserve_NAND_Blk 0x0033 000 000 000 Pre-fail
> Always - 16523
> 202 Percent_Lifetime_Used 0x0031 096 096 000 Pre-fail
> Offline - 4
> 206 Write_Error_Rate 0x000e 100 100 000 Old_age
> Always - 0
> 210 Success_RAIN_Recov_Cnt 0x0032 100 100 000 Old_age
> Always - 0
> 246 Total_Host_Sector_Write 0x0032 100 100 --- Old_age
> Always - 19329755627
> 247 Host_Program_Page_Count 0x0032 100 100 --- Old_age
> Always - 616802682
> 248 Bckgnd_Program_Page_Cnt 0x0032 100 100 --- Old_age
> Always - 3344366909
>
>
> Attribute 173 states that, on average, each block was erased 134 times,
> and states that it is about 4% of the liftime of the drive (attribute 202)
> was used. Check those/similar values on your drive. 4% on 8.611hs suggests
> that I can continue to use the drive another 23 more years.
>
> A few things you should consider:
>
> - There is no wear from reading
> - You should have as much free space as possible in order to level the
> wearing. If your filesystem is pretty much full, all the wear-leveing will
> occur on those few blocks that are released and re-used. If you only have
> one block left, you will kill it in 3000 writes :-)
> - Your filesystem shouls be aware that it is running on SSD in order
> to mark blocks as free (so the SSD can do it's magic). I finally dropped
> reiserfs for this reason.
> - lot more things to consider, but these are the most relevant IMHO
> - YMMV
>
>
> Would you like to share your counters? Filesystem? application?
>
> regards
>
> ariel
>
> El 22/07/18 a las 04:02, Jonathan Engwall escribió:
>
> Does 345 power_on_hours seem like "old age"
> Not happy. Thanks for the CEPH tip.
>
> On Sat, Jul 21, 2018, 11:30 PM Jonathan Engwall <
> engwalljonathanthereal at gmail.com> wrote:
>
>> I built up a couple r610 dells with the idea that they would boot from
>> ssd through a usb external drive. The cheap external drives might be the
>> real culprits. I have to think about that.
>> The latest drive to fail was a Kingston and I think it was new.
>> There is a write limit on some ssds?
>> Lately I have been building trying to crosscompile the cray xmp
>> simulator. It has thousands of targets. But earlier this week afterI
>> installed openvswitch was when the trouble began.
>> Jonathan Engwall
>>
>> On Sat, Jul 21, 2018, 10:44 PM John Hearns via Beowulf <
>> beowulf at beowulf.org> wrote:
>>
>>> I forgot the main purpose of the Internet. My bad.
>>> >You don't say for which purpose the SSDs are being used.
>>> Storing cat pics?
>>>
>>> On 22 July 2018 at 07:41, John Hearns <hearnsj at googlemail.com> wrote:
>>>
>>>> You don't say for which purpose the SSDs are being used.
>>>> In your laptop?
>>>> As system disks in HPC compute nodes?
>>>> Journalling drives in a parallel filesystem?
>>>> Data storage drives in a parallel filesystem?
>>>>
>>>> Over on the CEPH mailing list there are regular topics on choice of
>>>> SSDs. I would advise going over there and asking the same question.
>>>>
>>>> Also you don't say how they are failing.
>>>> Are these consumer grade drives or data centre grade drives?
>>>> Consumer drives have much, much lower 'drive writes per day'
>>>>
>>>> Two jobs ago I would have said the go-to data centre drives were Intel.
>>>>
>>>> Also one comment specific to HPC. I found that SSD drives dont 'get
>>>> sick' like spinning drives,
>>>> ie you dont see syslog messages about blocks failing to be
>>>> read/written. They just fail.
>>>> Also two jobs ago I though that SMART checks were not picking up
>>>> failing SSDs.
>>>> I believe that you can monitor them if you choose the correct counters.
>>>> Anyone?
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On 22 July 2018 at 07:32, Jonathan Aquilina <jaquilina at eagleeyet.net>
>>>> wrote:
>>>>
>>>>> What Ones are you currently getting?
>>>>>
>>>>> Sent from my iPhone
>>>>>
>>>>> > On 22 Jul 2018, at 05:42, Jonathan Engwall <
>>>>> engwalljonathanthereal at gmail.com> wrote:
>>>>> >
>>>>> > I am not happy with the SSDs I am using. I am buying another sad
>>>>> every couple weeks.
>>>>> > Fast sure, but my productivity right now is zero.
>>>>> > Are there any recommendations on reliable ssd brands?
>>>>> > Jonathan Engwall
>>>>> > _______________________________________________
>>>>> > Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin
>>>>> Computing
>>>>> > To change your subscription (digest mode or unsubscribe) visit
>>>>> http://www.beowulf.org/mailman/listinfo/beowulf
>>>>>
>>>>> _______________________________________________
>>>>> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin
>>>>> Computing
>>>>> To change your subscription (digest mode or unsubscribe) visit
>>>>> http://www.beowulf.org/mailman/listinfo/beowulf
>>>>>
>>>>
>>>>
>>> _______________________________________________
>>> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
>>> To change your subscription (digest mode or unsubscribe) visit
>>> http://www.beowulf.org/mailman/listinfo/beowulf
>>>
>>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20180722/07812347/attachment-0001.html>
More information about the Beowulf
mailing list