[zfs-discuss] very poor io performance

Gordan Bobic gordan.bobic at gmail.com
Wed Sep 25 05:04:42 EDT 2013


On Wed, Sep 25, 2013 at 9:59 AM, Zbigniew 'zibi' Jarosik <zibi at nora.pl>wrote:

> 2013/9/25 Gordan Bobic <gordan.bobic at gmail.com>
>
>> On Wed, Sep 25, 2013 at 9:28 AM, Zbigniew 'zibi' Jarosik <zibi at nora.pl>wrote:
>>
>>>
>>> 2013/9/24 Uncle Stoat <stoatwblr at gmail.com>
>>>
>>>> On 24/09/13 10:54, Zbigniew 'zibi' Jarosik wrote:
>>>>
>>>>> > ZIL is only going to be used for synchronous operations. It will not
>>>>> be
>>>>>
>>>>>> touched for anything else. For the sake of testing you can try setting
>>>>>> sync=disabled on the FS and see if that helps. If it doesn't, ZIL
>>>>>> won't
>>>>>> help either.
>>>>>>
>>>>>>  There is no way to force ZFS to use SSD as a cache for all writes?
>>>>>
>>>>
>>>> Firstly, you need to understand this statement:
>>>>
>>>> "ZIL is not a write cache. It is a write-intent store."
>>>>
>>>> In other words: ZIL is where pending writes are stored in case of
>>>> disaster.
>>>>
>>>
>>> Sorry, yesterday was a wery raw day ;/
>>>
>>> OK, that's obvious and explains everything, but was too complicated for
>>> me ;P
>>>
>>> So, another question - do you know any working ssd writeback solution
>>> working with ZOL? f.ex. maybe somebody tried out bcache?
>>>
>>
>> Have you actually established that it is the writes that are slow rather
>> than reads? Have you gathered the statistics on the read ops vs. write ops
>> during your typical workload?
>>
>
>
> Yes, i have >100GB of L2ARC on SSD - that covers much more than 150% of
> data that is actively readed, most of reads are from SSD (>90%) - i'm short
> on IOPS for writes in bursts.
>

If your workload is particularly bursty and the bursts aren't that big, you
may be better off investing in a decent batter-backed cache controller that
supports JBOD mode with write caching, and set it's read/write cache usage
to 100% writes.

To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20130925/e2b278cd/attachment.html>


More information about the zfs-discuss mailing list