apache pig - How to improve performance of PIG job that uses Datafu's Hyperloglog for estimating cardinality? -

-

i using datafu's hyperloglog udf estimate count of unique ids in dataset. in case have 320 million unique ids may appear multiple times in dataset.

dataset : country, id.

here code :

register datafu-1.2.0.jar;  define  hyperloglogplusplus datafu.pig.stats.hyperloglogplusplus();  -- id uuid, example : de305d54-75b4-431b-adb2-eb6b9e546014  all_ids = load '$data' using pigstorage(';') (country:chararray, id:chararray);  estimate_unique_ids = foreach (group all_ids country) generate     'total ids' label,     hyperloglogplusplus(all_ids) reach;  store estimate_unique_ids '$output' using pigstorage();

using 120 reducers noticed majority of them completed within minutes. handful of reducers overloaded data , ran forever. killed them after 24 hours.

i thought hyperloglog more efficient counting real. going wrong here?

in datafu 1.3.0, algebraic implementation of hyperloglog added. allows udf use combiner , improve performance in skewed situations.

however, in the comments in jira issue there discussion of other performance problems can arise when using hyperloglog. relevant quote below:

the thing keep in mind each instance of hyperloglogplus allocates pretty large byte array. can't remember exact numbers, think default precision of 20 hundreds of kb. in example if cardinality of "a" large going allocate lot of large byte arrays need transmitted combiner reducer. avoid using in "group by" situations unless know key cardinality quite small. udf better suited "group all" scenarios have lot of input data. if input data smaller byte array worse off using udf. if can accept worse precision byte array made smaller.


Comments

Post a Comment

Popular posts from this blog

php - Zend Framework / Skeleton-Application / Composer install issue -

-
i trying create zend framework application using skeleton in netbeans. when run composer install, following error message: "c:\php\php.exe" "c:\composer\composer.phar" "--ansi" "--no-interaction" "update" "--dev" using deprecated option "dev". dev packages installed default now. loading composer repositories package information updating dependencies (including require-dev) requirements not resolved installable set of packages. problem 1 - package requires php >=5.5 php version (5.4.42) not satisfy requirement. problem 2 - installation request zendframework/zendframework 2.5.1 -> satisfiable zendframework/zendframework[2.5.1]. - zendframework/zendframework 2.5.1 requires php >=5.5 -> php version (5.4.42) not satisfy requirement. done. apparently, zend framework only needs php 5.3+ contradicts above. need use php 5.4. my composer.json following: { "name": "
Read more

c# - Better 64-bit byte array hash -

-
i need hash algorithm produces 64-bit hash code ( long ) fewer collisions string.gethashcode() , fast (no expensive calls cryptographic functions). here's implementation of fnv still shows 3% of collisions after testing 2 million random strings. need number way lower. void main() { const string chars = "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz!#@$%^&*()_+}{\":?><,./;'[]0123456789\\"; const int n = 2000000; var random = new random(); var hashes = new hashset<long>(); int collisions = 0; for(int = 0; < n; i++) { var len = random.next(chars.length); var str = new char[len]; (int j = 0; j < len; j++) { str[j] = chars[random.next(chars.length)]; } var s = new string(str); if(!hashes.add(get64bithash( s ))) collisions++; } console.writeline("collision percentage after " + n + " random strings: " + ((doubl
Read more

python - PyCharm Type error Message -

-
i'm newbie trying out ml/dbn in python (3.4) under pycharm (comm.ed 4.5). following definition def make_thetas(xmin,xmax,n): xs = np.linespace(xmin,xmax,n) widths = (xs[1:] - xs[:-1])/2.0 thetas = xs[:-1] + widths return thetas throws error "class 'tuple' not define ' sub ', '-' operator cannot used on instance" on - operator on third line (widths = ....) ideas on how code running under pycharm - works alright in interactive python window. thx. to all, after g'g lot have found workaround seems work within pycharm. workaround because (even python newbie) expected python not need explicit typecasting, not when using datatypes imported lib such numpy. def make_thetas(xmin,xmax,n): xs = np.array(np.linspace(xmin,xmax,n)) widths = (xs[1:] - xs[:-1])/2.0 thetas = xs[:-1]+ widths return thetas using docstrings type-hinting such below did not work def make_thetas(xmin,xmax,n): "&q
Read more