With DirectQuerywithInMemory mode the queries use the relational data source by default, unless otherwise specified in the connection string from the client.
References:https://msdn.microsoft.com/en-us/library/hh230898(v=sql.120).aspx

Explanation:

By using SQL Profiler, you can intercept two classes of trace events from Analysis
Services, DAX Query Plan and DirectQuery events, both generated by the DirectQuery engine. Here, in this scenario we have a DAX Query.
DAX Query Plan events are generated by the DAX formula.
By using the In-Memory mode, you store a copy of data in the xVelocity (VertiPaq) storage engine.
Figure: This is how a query is executed by using In-Memory mode.

References: Microsoft SQL Server 2012 Analysis Services, The BISM Tabular Model,
Microsoft Press (July 2012), page 331
From Scenario: Users report that the query takes a long time to complete.

Explanation:

Step 1: Open the cube editor, and open the Dimension Usage tab.
Step 2: Configure a relationship between the Customer dimension and the Sales measure group. Use Day as the granularity.
From scenario: The SalesAnalysis cube contains a fact table named CoffeeSale loaded from a table named FactSale in the data warehouse. The time granularity within the cube is
15 minutes. The cube is processed every night at 23:00. You determine that the fact table cannot be fully processed in the expected time. Users have reported slow query response times.
Step 3: Reprocess the cube.
Step 4: Deploy the project changes.

Explanation:

Step 1: Partition the CoffeSale facto table.
Step 2: Set the storage mode for all partitions to HOLAP.
Partitions stored as HOLAP are smaller than the equivalent MOLAP partitions because they do not contain source data and respond faster than ROLAP partitions for queries involving summary data.
Step 3: Alter the processing job to ensure that it rearranges the partition structure each evening.
Step 4: Test that the cube meets the functional requirement for data currency and query performance.
From scenario:
Data analysts must be able to analyze sales for financial years, financial quarters, months, and days. Many reports are based on analyzing sales by month.
The SalesAnalysis cube contains a fact table named CoffeeSale loaded from a table named FactSale in the data warehouse. The time granularity within the cube is 15 minutes.
The cube is processed every night at 23:00. You determine that the fact table cannot be fully processed in the expected time. Users have reported slow query response times.
References:https://docs.microsoft.com/en-us/sql/analysis-services/multidimensional- models-olap-logical-cube-objects/partitions-partition-storage-modes-and-processing
Topic 3, Case Study #3
Background
You are a developer for a Seattle-based company. The company is expanding globally.
Many company employees speak fluent Mandarin and read Simplified Chinese.
You have six tabular data models that are deployed to two instances of Microsoft SQL
Server Analysis Services (SSAS).
Users report that the query takes a long time to complete.
You are planning the disk space allocations for a new Microsoft SQL Server Analysis
Services deployment. You plan to move several relational data file databases to the new
SSAS instance. The databases require a total of 10 GB of disk space.
You also plan to deploy Cubes and Aggregations and use Object Processing. Cubes will have small fact tables and few dimension members. No unnecessary aggregations will be created. You plan to process an entire cube in a single transaction.
Data Models
One of the data models is named CustomerSales. This data model contains eight tables.
The model includes a table named Sales that defines several measures, including a measure named PriorYearSales. The PriorYearSales measure is referenced by other measures, and is not intended to be analyzed directly by users. You must translate the metadata for all fata the CustomerSales data model to Simplifies Chinese. Team members from the Shanghai office assist with identifying appropriate translations.
A data model named OrderAnalysis is deployed to one of the SSAS instances. Order data is loaded into the OrderAnalysis data as part of an overnight process. You observe that the model is not up-to-date.
The business analysis team uses a variety of client applications to issue MDX queries against OrderAnalysis. Order data must be completely up-to-date.
The OrderAnalysis model has two user-defined hierarchies that are defined in a table named Order. New customers are only added once per day. The overnight process is sufficiently up-to-date for the Customer data to provide optimal performance while achieving the data currency goals whenever possible.
Databases
You deploy a database named DB1 to an SSAS instance as a project by using SQL Server
Data Tools. Data analysts report that they cannot access near real time data from the
SSAS SalesAnalysis model from DB1. You discover that the project has been deployed with the Direct Query Mode option set to OFF.
Most queries that use the SalesAnalysis data model use data from a table named
FactInternetSales that is 20 gigabyte (GB) in size. Cached data must be available for the
FactInternetSales table. All queries accessing the SalesAnalysis model must be executed in near real time.

With InMemory the queries can use the cache only.
References:https://msdn.microsoft.com/en-us/library/hh230898(v=sql.120).aspx