be like so I'll probably guess wrong.<br>
<br>
- Find some means of using SSDs as a form of tiered storage<br>
=A0=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D<br>
<br>
=A0We could continue deploying the majority of your storage from SATA<br>
=A0disks while also employing SSDs to cache these slower disks in<br>
=A0some manner.<br>
<br>
=A0The idea is that frequently-accessed data is backed on SSD whereas<br>
=A0data that is accessed less often is left on the larger-capacity<br>
=A0SATA, and *this remains transparent to the end user*, i.e. the VM.<br>
<br>
=A0This is not a new idea; plenty of storage hardware already does<br>
=A0it, ZFS can do it and so can BTRFS.<br>
<br>
=A0Advantages:<br>
<br>
=A0- For whatever benefit there is, everyone gets to feel it. If done<br>
=A0 =A0right, any VM that needs more IOPs should get more IOPs.<br>
<br>
=A0- Expensive SSDs purchased can be used immediately, in full.<br>
<br>
=A0Disadvantages:<br>
<br>
=A0- Since we can't use ZFS or expensive storage hardware, any<br>
=A0 =A0short-term solution is likely to be rather hacky. Do we want to<br>
=A0 =A0be pioneers here? This is your data.<br>
<br>
=A0- Customers with VMs that don't have heavy IO requirements (most)<b=
r>
=A0 =A0will be subsidising those who *do* have heavy IO requirements.<br>
=A0 =A0It's very unlikely we will put prices up, but SSDs are not free=
<br>
=A0 =A0so it has the effect of delaying the usual progression of<br>
=A0 =A0more-for-less that this type of service goes through.[1]<br>
<br>
=A0- Beyond what might be quite a blunt instrument, customers will<br>
=A0 =A0have no way to request faster storage and rely on it being<br>
=A0 =A0present. You have "some storage" and if that storage isn&=
#39;t<br>
=A0 =A0performing as fast as you would like, all we would be able to do<br=
>
=A0 =A0is try to see why it's not being cached on SSD.<br>
<br>
- Both?<br>
=A0=3D=3D=3D=3D=3D<br>
<br>
=A0Perhaps there is some way to do both? Maybe to start with using the<br>
=A0whole SSD as cache, but as requests to purchase SSD-backed storage<br>
=A0come in the cache could be reduced?<br>
<br>
=A0Advantages:<br>
<br>
=A0- Again everyone feels the benefit immediately and hardware isn't<b=
r>
=A0 =A0wasted.<br>
<br>
=A0- If the customer needs to buy SSD-backed storage then they can.<br>
<br>
=A0Disadvantages:<br>
<br>
=A0- If the caching is good enough then no one would feel the need to<br>
=A0 =A0buy SSD anyway, so why add complexity?<br>
<br>
=A0Questionable:<br>
<br>
=A0- If people buy all of the SSD, does that reduce caching benefit<br>
=A0 =A0to zero and suddenly screw everyone else over?<br>
<br>
=A0 =A0Presumably SSD-backed storage could be priced such that if a lot<br=
>
=A0 =A0of people did buy it, it would be economical to go out and buy a<br=
>
=A0 =A0pair of larger ones and swap them over without downtime[2].<br>
<br>
So, if anyone has any thoughts on this I'd be interested in hearing<br>
them.<br>
<br>
If you had an IO latency problem, would you know how to diagnose it<br>
to determine that it was something you were doing as opposed to<br>
"BitFolk's storage is overloaded but it's not me"?<br>
<br>
If you could do that, would you be likely to spend more money on<br>
SSD-backed storage?<br>
<br>
If we came to you and said that your VPS service was IO-bound and<br>
would run faster if you bought some SSD-backed storage, do you think<br>
that you would?[3]<br>
<br>
My gut feeling at the moment is that while I would love to be<br>
feeding the geek inside everyone and offering eleventy-billion<br>
choices, demand for SSD-backed storage at an additional cost will be<br>
low.<br>
<br>
I also think it's going to be very difficult for an admin of a<br>
virtualised block device to tell the difference between:<br>
<br>
=A0 =A0"All my processes are really slow at talking to storage; it=
9;s<br>
=A0 =A0because of my process ID 12345 which is a heavy DB query"<br>
<br>
and:<br>
<br>
=A0 =A0"All my processes are really slow at talking to storage; that&=
#39;s<br>
=A0 =A0definitely a problem with BitFolk's storage and not anything I<=
br>
=A0 =A0am doing."<br>
<br>
By the way, I think we've done reasonably well at keeping IO latency<br=
>
down, over the years:<br>
<br>
barbar: =A0 =A0<a href=3D"
http://tools.bitfolk.com/cacti/graphs/graph_1634_=
6.png" target=3D"_blank">
http://tools.bitfolk.com/cacti/graphs/graph_1634_6=
.png</a><br>
bellini: =A0 <a h
<b=%0A>%20r>%0A>%20%20=A0%20=A0will%20be%20subsidising%20those%20who%20*do*%20have%20heavy%20IO%20requirements.<br>%0A>%20%20=A0%20=A0It's%20very%20unlikely%20we%20will%20put%20prices%20up,%20but%20SSDs%20are%20not%20free=%0A>%20<br>%0A>%20%20=A0%20=A0so%20it%20has%20the%20effect%20of%20delaying%20the%20usual%20progression%20of<br>%0A>%20%20=A0%20=A0more-for-less%20that%20this%20type%20of%20service%20goes%20through.%5B1%5D<br>%0A>%20<br>%0A>%20%20=A0-%20Beyond%20what%20might%20be%20quite%20a%20blunt%20instrument,%20customers%20will<br>%0A>%20%20=A0%20=A0have%20no%20way%20to%20request%20faster%20storage%20and%20rely%20on%20it%20being<br>%0A>%20%20=A0%20=A0present.%20You%20have%20"some%20storage"%20and%20if%20that%20storage%20isn&=%0A>%20#39;t<br>%0A>%20%20=A0%20=A0performing%20as%20fast%20as%20you%20would%20like,%20all%20we%20would%20be%20able%20to%20do<br=%0A>%20>%0A>%20%20=A0%20=A0is%20try%20to%20see%20why%20it's%20not%20being%20cached%20on%20SSD.<br>%0A>%20<br>%0A>%20-%20Both?<br>%0A>%20%20=A0=3D=3D=3D=3D=3D<br>%0A>%20<br>%0A>%20%20=A0Perhaps%20there%20is%20some%20way%20to%20do%20both?%20Maybe%20to%20start%20with%20using%20the<br>%0A>%20%20=A0whole%20SSD%20as%20cache,%20but%20as%20requests%20to%20purchase%20SSD-backed%20storage<br>%0A>%20%20=A0come%20in%20the%20cache%20could%20be%20reduced?<br>%0A>%20<br>%0A>%20%20=A0Advantages:<br>%0A>%20<br>%0A>%20%20=A0-%20Again%20everyone%20feels%20the%20benefit%20immediately%20and%20hardware%20isn't<b=%0A>%20r>%0A>%20%20=A0%20=A0wasted.<br>%0A>%20<br>%0A>%20%20=A0-%20If%20the%20customer%20needs%20to%20buy%20SSD-backed%20storage%20then%20they%20can.<br>%0A>%20<br>%0A>%20%20=A0Disadvantages:<br>%0A>%20<br>%0A>%20%20=A0-%20If%20the%20caching%20is%20good%20enough%20then%20no%20one%20would%20feel%20the%20need%20to<br>%0A>%20%20=A0%20=A0buy%20SSD%20anyway,%20so%20why%20add%20complexity?<br>%0A>%20<br>%0A>%20%20=A0Questionable:<br>%0A>%20<br>%0A>%20%20=A0-%20If%20people%20buy%20all%20of%20the%20SSD,%20does%20that%20reduce%20caching%20benefit<br>%0A>%20%20=A0%20=A0to%20zero%20and%20suddenly%20screw%20everyone%20else%20over?<br>%0A>%20<br>%0A>%20%20=A0%20=A0Presumably%20SSD-backed%20storage%20could%20be%20priced%20such%20that%20if%20a%20lot<br=%0A>%20>%0A>%20%20=A0%20=A0of%20people%20did%20buy%20it,%20it%20would%20be%20economical%20to%20go%20out%20and%20buy%20a<br=%0A>%20>%0A>%20%20=A0%20=A0pair%20of%20larger%20ones%20and%20swap%20them%20over%20without%20downtime%5B2%5D.<br>%0A>%20<br>%0A>%20So,%20if%20anyone%20has%20any%20thoughts%20on%20this%20I'd%20be%20interested%20in%20hearing<br>%0A>%20them.<br>%0A>%20<br>%0A>%20If%20you%20had%20an%20IO%20latency%20problem,%20would%20you%20know%20how%20to%20diagnose%20it<br>%0A>%20to%20determine%20that%20it%20was%20something%20you%20were%20doing%20as%20opposed%20to<br>%0A>%20"BitFolk's%20storage%20is%20overloaded%20but%20it's%20not%20me"?<br>%0A>%20<br>%0A>%20If%20you%20could%20do%20that,%20would%20you%20be%20likely%20to%20spend%20more%20money%20on<br>%0A>%20SSD-backed%20storage?<br>%0A>%20<br>%0A>%20If%20we%20came%20to%20you%20and%20said%20that%20your%20VPS%20service%20was%20IO-bound%20and<br>%0A>%20would%20run%20faster%20if%20you%20bought%20some%20SSD-backed%20storage,%20do%20you%20think<br>%0A>%20that%20you%20would?%5B3%5D<br>%0A>%20<br>%0A>%20My%20gut%20feeling%20at%20the%20moment%20is%20that%20while%20I%20would%20love%20to%20be<br>%0A>%20feeding%20the%20geek%20inside%20everyone%20and%20offering%20eleventy-billion<br>%0A>%20choices,%20demand%20for%20SSD-backed%20storage%20at%20an%20additional%20cost%20will%20be<br>%0A>%20low.<br>%0A>%20<br>%0A>%20I%20also%20think%20it's%20going%20to%20be%20very%20difficult%20for%20an%20admin%20of%20a<br>%0A>%20virtualised%20block%20device%20to%20tell%20the%20difference%20between:<br>%0A>%20<br>%0A>%20%20=A0%20=A0"All%20my%20processes%20are%20really%20slow%20at%20talking%20to%20storage;%20it=%0A>%209;s<br>%0A>%20%20=A0%20=A0because%20of%20my%20process%20ID%2012345%20which%20is%20a%20heavy%20DB%20query"<br>%0A>%20<br>%0A>%20and:<br>%0A>%20<br>%0A>%20%20=A0%20=A0"All%20my%20processes%20are%20really%20slow%20at%20talking%20to%20storage;%20that&=%0A>%20#39;s<br>%0A>%20%20=A0%20=A0definitely%20a%20problem%20with%20BitFolk's%20storage%20and%20not%20anything%20I<=%0A>%20br>%0A>%20%20=A0%20=A0am%20doing."<br>%0A>%20<br>%0A>%20By%20the%20way,%20I%20think%20we've%20done%20reasonably%20well%20at%20keeping%20IO%20latency<br=%0A>%20>%0A>%20down,%20over%20the%20years:<br>%0A>%20<br>%0A>%20barbar:%20=A0%20=A0<a%20href=3D%22http://tools.bitfolk.com/cacti/graphs/graph_1634_=%0A>%206.png%22%20target=3D%22_blank%22>http://tools.bitfolk.com/cacti/graphs/graph_1634_6=%0A>%20.png</a><br>%0A>%20bellini:%20=A0%20<a%20h "Reply to this message")
