Reminiscence Optimizations for Analytic Queries in Cloudera Information Warehouse

Apache Impala is used at this time by over 1,000 clients to energy their analytics in on premise in addition to cloud-based deployments. Massive person communities of analysts and builders profit from Impala’s quick question execution, serving to them get their work finished extra successfully. For these customers efficiency and concurrency are at all times high of thoughts. 

An necessary method to make sure good efficiency and concurrency is thru environment friendly utilization of reminiscence. If we are able to make higher use of reminiscence, much less time is spent with queries queueing up ready at no cost reminiscence, so outcomes come again quicker. Equally, with higher utilization of obtainable reminiscence extra customers can question the info at any given time, so extra individuals can use the warehouse on the similar time. The top outcome – happier customers, and extra of them.

This publish explains the novel method for a way Impala, supplied throughout the Cloudera Information Platform (CDP), is now capable of get rather more mileage out of the reminiscence at its disposal.

Impala has at all times centered on effectivity and pace, being written in C++ and successfully utilizing strategies similar to runtime code technology and multithreading. You possibly can learn earlier weblog posts on Impala’s efficiency and querying strategies right here – “New Multithreading Mannequin for Apache Impala”, “Preserving Small Queries Quick – Brief question optimizations in Apache Impala” and “Quicker Efficiency for Selective Queries”. 

Analytical SQL workloads use aggregates and joins closely. Therefore, optimizing such operators for each efficiency and effectivity in analytical engines like Impala will be very helpful. We are going to now look into one of many strategies used to scale back peak reminiscence utilization of Aggregates and Joins by as much as 50%, and peak node reminiscence utilization by 18% on the per-node degree on a TPC-DS 10000 workload.

Hash Desk

Each Aggregates and Joins in Impala use a Hash Desk, and we are going to present how we lowered its measurement for the operation. The HashTable class implementation in Impala includes a contiguous array of Bucket, and every Bucket incorporates both a pointer to knowledge or a pointer to a linked record of duplicate entries named DuplicateNode.

These are the buildings of Bucket and DuplicateNode (with a couple of particulars modified for simplicity):

struct DuplicateNode {

    bool matched; // 1-byte

    // padding of 7-bytes

    DuplicateNode* subsequent; // 8-byte pointer to subsequent DuplicateNode

    Information* knowledge; // 8-byte pointer to knowledge being hashed


  struct Bucket {

    bool crammed; // 1-byte

    bool matched; // 1-byte

    bool hasDuplicates; // 1-byte

    // padding of 1-byte

    uint32_t hash; // 4-byte

    // bucketData is a pointer to DuplicateNode or

    // pointer to Information.

    union {

      Information* knowledge; // pointer to knowledge being hashed

      DuplicateNode* duplicates;

    } bucketData; // 8-byte


When evaluating the scale for a struct, these are a number of the guidelines for reminiscence alignment, assuming 64-bit system:

  1. Reminiscence tackle for particular person members begins at reminiscence tackle divisible by its measurement. So a pointer will begin at reminiscence divisible by 8, a bool by 1 and uint32_t by 4. Members can be preceded by padding if wanted to verify the beginning tackle is divisible by their measurement.
  2. Measurement of struct can be aligned to it’s largest member. As an example, in each the structs above the biggest member is a pointer of measurement 8 bytes. Therefore, the scale of struct can be a a number of of 8 too.

Primarily based on the above guidelines,  Bucket within the above snippet is commented with measurement occupied by each member and padding wherever required. Complete measurement of the Bucket is 16 bytes. Equally, the overall measurement of DuplicateNode is 24 bytes.

We determined to scale back the scale of the Bucket and DuplicateNode by eradicating bool fields from each, decreasing the sizes to 12 bytes and 16 bytes respectively. Nonetheless 12 bytes isn’t a legitimate measurement of Bucket, because it must be a a number of of 8 bytes (the scale of the biggest member of the struct). In such circumstances, we are able to use __attribute__ ((packed)) to make sure struct packing in order that the scale is 12 bytes.

How can we obtain eradicating these booleans, as they have to be current for each Bucket and DuplicateNode?

tl;dr: We determined to take away all bool members by folding them right into a pointer that’s already a part of the struct.

Folding knowledge into pointers

Intel Stage 5 proposal 64-bit reminiscence tackle

On 64-bit architectures, pointers retailer reminiscence addresses utilizing 8 bytes. However on architectures like x86 and ARM the linear tackle is proscribed to 48 bits lengthy, with bits 49 to 64 reserved for future utilization. Sooner or later with Intel’s degree 5 paging proposal (whitepaper), it’s planning to chill out the restrict to 57-bit on x86, which implies we are able to use probably the most vital 7 bits – i.e. bits 58 to 64 – to retailer further knowledge. One caveat is that even when simply 48 bits out of 64 bits are wanted to learn a reminiscence, the processor checks if vital bits (48…64) are equivalent – i.e. signal prolonged. If not, such an tackle will trigger a fault. It means folded pointers could not at all times be storing a legitimate addressable reminiscence. Therefore folded pointers have to be signal prolonged earlier than dereferencing.

We use the above method to fold crammed, matched and hasDuplicates into pointer bucketData. After folding and struct packing we are going to get a Bucket measurement of 12 bytes. Equally DuplicateNode will be lowered to 16 bytes as an alternative of 24 bytes. In whole, we cut back the reminiscence necessities for these two structs from 40 bytes to twenty-eight bytes, a discount of 30%.

Different necessities

In our implementation, there’s a requirement relating to the scale of Bucket and variety of buckets in hash desk to be an influence of two. These necessities are for the next causes:

  1. Inner reminiscence allocator allocates reminiscence in energy of two to keep away from inner fragmentation. Therefore, variety of buckets * sizeof(Bucket) needs to be an influence of two.
  2. Variety of buckets (‘N’) being the ability of two allows quicker modulo operations.

As an alternative of utilizing a sluggish modulo operation (hash % N), quicker bitwise operation (hash & (N-1)) can be utilized when N is energy of two.

As a result of this, a 4 bytes hash subject from Bucket is eliminated and saved individually in a brand new array hash_array_ in HashTable class. This ensures sizeof(Bucket) is 8 which is energy of two. One other benefit of separating hash is that Bucket isn’t required to be packed now.

Experimental analysis:

We did in depth analysis of the method to see the way it impacts efficiency and reminiscence utilization. We used 3 benchmarks:

  1. Microbenchmark: We ran the construct and probe strategies 60 occasions on a smaller variety of rows to judge the efficiency and reminiscence consumed.
  2. Billion-Row benchmark: On a single daemon, we ran the construct and probe benchmark for a billion rows to measure the efficiency and reminiscence consumed.
  3. TPC-DS-10000: Total TPC-DS benchmark of scale 10000 was run on a 17-node cluster to measure the efficiency. It additionally measured peak reminiscence consumed on the node and the operator degree.


Determine 2a. Reminiscence benchmark

Determine 2a exhibits the outcomes of the reminiscence benchmark. Benchmark names are within the format memory_XX_YY, the place XX is the variety of values being inserted into the hash desk and YY represents the share of distinctive values. We see a discount in reminiscence consumed by as much as 30% on constructing the hash desk.

Determine 2b. Runtime benchmark

Determine 2b exhibits the outcomes of the efficiency benchmark. build_XX_YY represents the construct benchmark, the place XX values have been inserted and YY is the share of distinctive values. Equally probe_XX_YY would probe towards a hash desk constructed with XX rows and YY distinctive values. These benchmarks have been run 60 occasions, and so they have been repeated 10 occasions to search out out iterations per millisecond. Determine 2b exhibits the ninetieth percentile of the variety of iterations measured for these 60 runs. We observe no vital distinction within the runtime of those hash desk operations as a result of this alteration. 

Billion-Row benchmark

We used the TPC-DS gross sales and gadgets desk for this benchmark. gross sales had columns s_item_id (int), s_quantity(int) ,s_date(date), whereas gadgets had columns i_item_id (int)and i_price (double). gross sales had 1 billion rows and gadgets had 30 million rows. 

Construct Benchmark

We ran a Group By question on gross sales to measure the efficiency and reminiscence of constructing a hash desk.

Question: choose depend(*) from gross sales group by s_item_id having depend(*) > 9999999999;

Grouping Combination Reminiscence Utilization
With Modifications With out Modifications
Peak Allocation Cumulative Allocation Peak Allocation Cumulative Allocation
1.14G 1.85G 1.38G 2.36GB

Determine 3a

As proven in Determine 3a, we noticed peak allocation lowered by 17% and cumulative allocation lowered by

21%. When working this 20 occasions, we didn’t see any efficiency degradation. Geomean with adjustments and with out adjustments have been round 68 seconds in each circumstances.

Probe Benchmark

For measuring the probe we ran a be part of question between gadgets and gross sales, the place gross sales is on the probe aspect and gadgets is on the construct aspect. Since we’re constructing a hash desk solely on a smaller desk within the be part of proposed, the objective of this benchmark was to not measure the discount in reminiscence, however to measure any efficiency distinction in probing 1 billion rows through the gross sales desk.

Nonetheless, we created 3 sorts of gross sales tables for this function:

  1. sales_base: It has randomly generated 1 billion rows, the identical that was used within the Construct benchmarks.
  2. sales_30: It has 1 billion rows, with 30% of the rows distinctive.
  3. sales_60: It has 1 billion rows, with 60% of the rows distinctive.

We noticed comparable efficiency in each runs with our adjustments being barely quicker on sales_base, as proven in Determine 3b. Subsequently, whereas decreasing reminiscence consumption we didn’t measure any degradation in mixture question runtime.

Desk Sort GEOMEAN over 20 runs (seconds)
With adjustments With out adjustments
gross sales 110.8551081 114.6912898
sales_30 103.2863058 102.4787489
sales_60 84.12813181 84.8765098

Determine 3b

TPCDS-10000 scale

We evaluated the brand new hash implementation towards a TPC-DS workload at scale 10000. We ran all of the workload queries on a 17 node cluster with knowledge saved in HDFS.

Determine 4a

Per-Operator Discount:

For each question we computed the utmost proportion of discount in reminiscence for particular person Be a part of and Aggregation operators. We thought-about solely the operators larger than 10 MB. As per Determine 4a, we discovered that for 42 out of 99 queries, reminiscence consumption lowered by greater than 10%. Moreover, for twenty-four of these queries we noticed reminiscence consumption lowered by greater than 20%.

Per-Node reminiscence discount:

On computing common peak reminiscence consumption for the nodes concerned, 28 queries confirmed larger than 5% discount in reminiscence and 11 queries confirmed greater than 10% discount, as seen in Determine 4b. Moreover, we noticed round a most of 18% discount, for q72.

Determine 4b

Determine 4c

On contemplating max-peak reminiscence consumed in any node for a question, 27 queries present a discount by 5%, and 11 present discount greater than 10%, as seen in Determine 4c. The utmost discount noticed was greater than 20%, for q65.


As proven within the earlier part we noticed vital discount in reminiscence each on the node degree and operator degree, with none efficiency degradation. 

This reminiscence effectivity and efficiency optimization, in addition to many others in Impala, is what makes it the popular alternative for enterprise intelligence and analytics workloads, particularly at scale. Now that increasingly knowledge warehousing is finished within the cloud, a lot of that within the Cloudera Information Warehouse knowledge service, efficiency enchancment straight equates to value financial savings. The quicker the queries run, the earlier the assets will be launched so the person not pays for them. A latest benchmark by a 3rd get together exhibits how Cloudera has the perfect price-performance on the cloud knowledge warehouse market.

We encourage everybody to take a tour or take a look at drive Apache Impala throughout the Cloudera Information Warehouse knowledge service to see the way it performs on your workloads. You can even contact your gross sales consultant to e-book a demo. Moreover, to have interaction with the broader group please join at [email protected] or [email protected].

Leave a Comment