hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From kisalay <kisa...@gmail.com>
Subject Re: Add client complexity or use a coprocessor?
Date Wed, 11 Apr 2012 19:32:30 GMT
Assuming that there are X records per region, the performance will be
determined by:

scan time required for X records + Time required to update the bitset
counter for X records.

I will not be very amazed if updating the bitset dominates the cost
here, as for each record you may have to generate one or many hash
values. This really depends on the implementation of the counting
algorithm.

But then you can optimize this by precomputing the hash value / bit
position to be set in the bitset at insertion time itself.



On Wed, Apr 11, 2012 at 11:11 PM, Tom Brown <tombrown52@gmail.com> wrote:
> kisalay,
>
> Are you talking about storing all my data in a non-aggregate format,
> and just aggregating as needed?  If so, do you have any idea what kind
> of performance I should expect when scanning over 15 million rows to
> summarize the specific cubes I need (using bitfields to estimate
> unique users on each region and merge them together later)?
>
> Or are you suggesting something else?
>
> --Tom
>
> On Tue, Apr 10, 2012 at 11:59 PM, kisalay <kisalay@gmail.com> wrote:
>> Tom,
>>
>> I was way too curious to resist a reply here.
>> If you want to store a bytearray estimating the unique count for a
>> particular OLAP cell, will you not see a lot of updates to the same
>> cell and create a hotspot ?
>>
>> I think another option comes to my mind. I assume that you get all the
>> user activities. Now consider for a moment that you store each of
>> there activities in you Table in HBase with the row-key being
>> udid-timestamp and value being some blob representing the activity
>> detail. Now if I have to do a unique count of user for a OLAP cell, I
>> would create endpoint coprocessors, that would execute per region,
>> scan it, and prepare a bitset representing the unique counts for the
>> OLAP cellof interest for that region.
>>
>> Now if you are implementing Probabilistic Counting, you can OR the
>> bitsets returned from each region to get the final bitset which will
>> give you the overall unique counts for all the regions together.
>>
>> You would not only save on network transfers, as you are doing the
>> counts per region in coprocessor and returning only one bitset per
>> region, you would also be able to resolve the query in the time taken
>> by one coprocessor to scan one region.
>>
>> I have a rudimentary implementation of Probabilistic Count which I
>> once used as a Bolt in Storm (Storm was developed at Backtype and
>> opensourced by Twitter) to count the unique users for a similar use
>> case. Let me know if you would to look at the implementation of the
>> algorithm.
>>
>>
>>
>> ~Kisalay
>>
>> On Wed, Apr 11, 2012 at 5:23 AM, Andrew Purtell <apurtell@apache.org> wrote:
>>>> Even my implementation of an atomic increment
>>>> (using a coprocessor) is two orders of magnitude slower than the
>>>> provided implementation.  Are there properties inherent to
>>>> coprocessors or Incrementors that would force this kind of performance
>>>> difference?
>>>
>>>
>>> No.
>>>
>>>
>>> You may be seeing a performance difference if you are packing multiple Increments
into one round trip but not doing a similar kind of batching if calling a custom endpoint.
Each Endpoint invocation is a round trip unless you do something like:
>>>
>>>     List<Row> actions = new ArrayList<Row>();    actions.add(new
Exec(conf, row, protocol, method, ...));
>>>
>>>     actions.add(new Exec(conf, row, protocol, method, ...));
>>>
>>>     actions.add(new Exec(conf, row, protocol, method, ...));
>>>
>>>     Object[] results = table.batch(actions);
>>>     ...
>>>
>>>
>>> I've not personally tried that particular API combination but don't see why it
would not be possible.
>>>
>>>
>>> Beyond that, I'd suggest running a regionserver with your coprocessor installed
under a profiler to see if you have monitor contention or a hotspot or similar. It could be
something unexpected.
>>>
>>>
>>>> Can you think of an efficient way to implement an atomic bitfield
>>>> (other than adding it as a separate feature like atomic increments)?
>>>
>>> I think the idea of an atomic bitfield operation as part of the core API is intriguing.
It has applicability to your estimator use case and I can think of a couple of things I could
use it for. If there is more support for this idea, this may be something to consider.
>>>
>>>
>>> Best regards,
>>>
>>>
>>>     - Andy
>>>
>>> Problems worthy of attack prove their worth by hitting back. - Piet Hein (via
Tom White)
>>>
>>>
>>>
>>> ----- Original Message -----
>>>> From: Tom Brown <tombrown52@gmail.com>
>>>> To: user@hbase.apache.org; Andrew Purtell <apurtell@apache.org>
>>>> Cc:
>>>> Sent: Tuesday, April 10, 2012 3:53 PM
>>>> Subject: Re: Add client complexity or use a coprocessor?
>>>>
>>>> Andy,
>>>>
>>>> I have attempted to use coprocessors to achieve a passable performance
>>>> but have failed so far. Even my implementation of an atomic increment
>>>> (using a coprocessor) is two orders of magnitude slower than the
>>>> provided implementation.  Are there properties inherent to
>>>> coprocessors or Incrementors that would force this kind of performance
>>>> difference?
>>>>
>>>> Can you think of an efficient way to implement an atomic bitfield
>>>> (other than adding it as a separate feature like atomic increments)?
>>>>
>>>> Thanks!
>>>>
>>>> --Tom
>>>>
>>>> On Tue, Apr 10, 2012 at 12:01 PM, Andrew Purtell <apurtell@apache.org>
>>>> wrote:
>>>>>  Tom,
>>>>>>  I am a big fan of the Increment class. Unfortunately, I'm not doing
>>>>>>  simple increments for the viewer count. I will be receiving duplicate
>>>>>>  messages from a particular client for a specific cube cell, and
>>>> don't
>>>>>>  want them to be counted twice
>>>>>
>>>>>  Gotcha.
>>>>>
>>>>>>  I created an RPC endpoint coprocessor to perform this function
but
>>>>>>  performance suffered heavily under load (it appears that the endpoint
>>>>>>  performs all functions in serial).
>>>>>
>>>>>  Did you serialize access to your data structure(s)?
>>>>>
>>>>>>  When I tried implementing it as a region observer, I was unsure
of how
>>>>>>  to correctly replace the provided "put" with my own. When I
>>>> issued a
>>>>>>  put from within "prePut", the server blocked the new put
>>>> (waiting for
>>>>>>  the "prePut" to finish). Should I be attempting to modify the
>>>> WALEdit
>>>>>>  object?
>>>>>
>>>>>  You can add KVs to the WALEdit. Or, you can get a reference to the
>>>> Put's familyMap:
>>>>>
>>>>>      Map<byte[], List<KeyValue>> familyMap = put.getFamilyMap();
>>>>>
>>>>>  and if you modify the map, you'll change what gets committed.
>>>>>
>>>>>>  Is there a way to extend the functionality of "Increment" to
>>>> provide
>>>>>>  arbitrary bitwise operations on a the contents of a field?
>>>>>
>>>>>  As a matter of design, this should be a new operation. It does sound
>>>> interesting and useful, some sort of atomic bitfield.
>>>>>
>>>>>
>>>>>  Best regards,
>>>>>
>>>>>      - Andy
>>>>>
>>>>>  Problems worthy of attack prove their worth by hitting back. - Piet
Hein
>>>> (via Tom White)
>>>>>
>>>>>
>>>>>
>>>>>  ----- Original Message -----
>>>>>>  From: Tom Brown <tombrown52@gmail.com>
>>>>>>  To: user@hbase.apache.org
>>>>>>  Cc:
>>>>>>  Sent: Monday, April 9, 2012 10:14 PM
>>>>>>  Subject: Re: Add client complexity or use a coprocessor?
>>>>>>
>>>>>>  Andy,
>>>>>>
>>>>>>  I am a big fan of the Increment class. Unfortunately, I'm not doing
>>>>>>  simple increments for the viewer count. I will be receiving duplicate
>>>>>>  messages from a particular client for a specific cube cell, and
>>>> don't
>>>>>>  want them to be counted twice (my stats don't have to be 100%
>>>>>>  accurate, but the expected rate of duplicates will be higher than
the
>>>>>>  allowable error rate).
>>>>>>
>>>>>>  I created an RPC endpoint coprocessor to perform this function
but
>>>>>>  performance suffered heavily under load (it appears that the endpoint
>>>>>>  performs all functions in serial).
>>>>>>
>>>>>>  When I tried implementing it as a region observer, I was unsure
of how
>>>>>>  to correctly replace the provided "put" with my own. When I
>>>> issued a
>>>>>>  put from within "prePut", the server blocked the new put
>>>> (waiting for
>>>>>>  the "prePut" to finish). Should I be attempting to modify the
>>>> WALEdit
>>>>>>  object?
>>>>>>
>>>>>>  Is there a way to extend the functionality of "Increment" to
>>>> provide
>>>>>>  arbitrary bitwise operations on a the contents of a field?
>>>>>>
>>>>>>  Thanks again!
>>>>>>
>>>>>>  --Tom
>>>>>>
>>>>>>>  If it helps, yes this is possible:
>>>>>>>
>>>>>>>>   Can I observe updates to a
>>>>>>>>   particular table and replace the provided data with my
own?
>>>> (The
>>>>>>>>   client calls "put" with the actual user ID, my
>>>> co-processor
>>>>>>  replaces
>>>>>>>>   it with a computed value, so the actual user ID never
gets
>>>> stored in
>>>>>>>>   HBase).
>>>>>>>
>>>>>>>  Since your option #2 requires atomic updates to the data structure,
>>>> have you
>>>>>>  considered native
>>>>>>>  atomic increments? See
>>>>>>>
>>>>>>>
>>>> http://hbase.apache.org/apidocs/org/apache/hadoop/hbase/client/HTable.html#incrementColumnValue%28byte[],%20byte[],%20byte[],%20long,%20boolean%29
>>>>>>>
>>>>>>>
>>>>>>>  or
>>>>>>>
>>>>>>>
>>>> http://hbase.apache.org/apidocs/org/apache/hadoop/hbase/client/Increment.html
>>>>>>>
>>>>>>>  The former is a round trip for each value update. The latter
allows
>>>> you to
>>>>>>  pack multiple updates
>>>>>>>  into a single round trip. This would give you accurate counts
even
>>>> with
>>>>>>  concurrent writers.
>>>>>>>
>>>>>>>  It should be possible for you to do partial aggregation on
the
>>>> client side
>>>>>>  too whenever parallel
>>>>>>>  requests colocate multiple updates to the same cube within
some
>>>> small window
>>>>>>  of time.
>>>>>>>
>>>>>>>  Best regards,
>>>>>>>
>>>>>>>
>>>>>>>      - Andy
>>>>>>>
>>>>>>>  Problems worthy of attack prove their worth by hitting back.
- Piet
>>>> Hein
>>>>>>  (via Tom White)
>>>>>>>
>>>>>>>  ----- Original Message -----
>>>>>>>>   From: Tom Brown <tombrown52@gmail.com>
>>>>>>>>   To: user@hbase.apache.org
>>>>>>>>   Cc:
>>>>>>>>   Sent: Monday, April 9, 2012 9:48 AM
>>>>>>>>   Subject: Add client complexity or use a coprocessor?
>>>>>>>>
>>>>>>>>   To whom it may concern,
>>>>>>>>
>>>>>>>>   Ignoring the complexities of gathering the data, assume
that I
>>>> will be
>>>>>>>>   tracking millions of unique viewers. Updates from each
of our
>>>> millions
>>>>>>>>   of clients are gathered in a centralized platform and
spread
>>>> among a
>>>>>>>>   group of machines for processing and inserting into HBase
>>>> (assume that
>>>>>>>>   this group can be scaled horizontally). The data is stored
in
>>>> an OLAP
>>>>>>>>   cube format and one of the metrics I'm tracking across
>>>> various
>>>>>>>>   attributes is viewership (how many people from Y are
watching
>>>> X).
>>>>>>>>
>>>>>>>>   I'm writing this to ask for your thoughts as to the most
>>>>>>  appropriate
>>>>>>>>   way to structure my data so I can count unique TV viewers
>>>> (assume a
>>>>>>>>   service like netflix or hulu).
>>>>>>>>
>>>>>>>>   Here are the solutions I'm considering:
>>>>>>>>
>>>>>>>>   1. Store each unique user ID as the cell name within
the
>>>> cube(s) it
>>>>>>>>   occurs. This has the advantage of having 100% accuracy,
but
>>>> the
>>>>>>>>   downside is the enormous space required to store each
unique
>>>> cell.
>>>>>>>>   Consuming this data is also problematic as the only way
to
>>>> provide a
>>>>>>>>   viewership count is by counting each cell. To save the
>>>> overhead of
>>>>>>>>   sending each cell over the network, counting them could
be
>>>> done by a
>>>>>>>>   coprocessor on the region server, but that still doesn't
>>>> avoid the
>>>>>>>>   overhead of reading each cell from the disk. I'm also
not
>>>> sure what
>>>>>>>>   happens if a single row is larger than an entire region
(48
>>>> bytes per
>>>>>>>>   user ID * 10,000,000 users = 480GB).
>>>>>>>>
>>>>>>>>   2. Store a byte array that allows estimating unique viewers
>>>> (with a
>>>>>>>>   small margin of error*). Add a co-processor for updating
this
>>>> column
>>>>>>>>   so I can guarantee the updates to a specific OLAP cell
will be
>>>> atomic.
>>>>>>>>   The main benefit from this path is that there the nodes
that
>>>> update
>>>>>>>>   HBase can be less complex. Another benefit I see is that
the I
>>>> can
>>>>>>>>   just add more HBase regions as scale requires. However,
>>>> I'm not
>>>>>>  sure
>>>>>>>>   if I can use a coprocessor the way I want; Can I observe
>>>> updates to a
>>>>>>>>   particular table and replace the provided data with my
own?
>>>> (The
>>>>>>>>   client calls "put" with the actual user ID, my
>>>> co-processor
>>>>>>  replaces
>>>>>>>>   it with a computed value, so the actual user ID never
gets
>>>> stored in
>>>>>>>>   HBase).
>>>>>>>>
>>>>>>>>   3. Store a byte array that allows estimating unique viewers
>>>> (with a
>>>>>>>>   small margin of error*). Re-arrange my architecture so
that
>>>> each OLAP
>>>>>>>>   cell is only updated by a single node. The main benefit
from
>>>> this
>>>>>>>>   would be that I don't need to worry about atomic
>>>> operations in
>>>>>>  HBase
>>>>>>>>   since all updates for a single cell will be atomic and
in
>>>> serial. The
>>>>>>>>   biggest downside is that I believe it will add significant
>>>> complexity
>>>>>>>>   to my overall architecture.
>>>>>>>>
>>>>>>>>
>>>>>>>>   Thanks for your time, and I look forward to hearing your
>>>> thoughts.
>>>>>>>>
>>>>>>>>   Sincerely,
>>>>>>>>   Tom Brown
>>>>>>>>
>>>>>>>>   *(For information about the byte array mentioned in #2
and #3,
>>>> see:
>>>>>>>>
>>>>>>
>>>> http://highscalability.com/blog/2012/4/5/big-data-counting-how-to-count-a-billion-distinct-objects-us.html)
>>>>>>>>
>>>>>>
>>>>

Mime
View raw message