Paper Summary: A Self-Configurable Geo-Replicated Cloud Storage System
the Pileus work, which I had covered here. Pileus aimed to assistance developers give away a suitable consistency/latency combination for their application deployment. In Pileus, the configuration of original together with secondary nodes is assumed to hold upwardly fixed (some storage nodes are designated every bit original nodes, which concur the original data, piece others are secondary nodes). The developer uses an SLA to Blue Planet ranked preferences for latency together with consistency of the reads that would brand nigh feel for the application, together with using this SLA, Pileus provides dynamic tuning of the performance of the application past times deciding which read to frontward to which replica/master.
This newspaper introduces a followup system, called Tuba, where the configuration is non fixed, together with tin hold upwardly changed on the fly. Tuba extends Pileus to address the occupation of finding an optimal configuration of original together with secondary replicas that maximizes the overall utility together with minimizes the toll for the application.
Tuba extends Pileus amongst a configuration service (CS) delivering the next capabilities:
1. performing a reconfiguration periodically for dissimilar tablets, and
2. informing clients of the electrical flow configuration for dissimilar tablets.
Data is organized into tablets every bit access units, together with each tablet is assigned to original together with secondary nodes. Of course of written report dissimilar tablets may hold upwardly assigned to dissimilar original together with secondary nodes. In guild for the CS to configure a tablet's replicas to maximize the overall utility, the CS must hold upwardly aware of the means the tablet is beingness accessed globally. Therefore, all clients inward the organization periodically post their observed latency together with the hitting together with missy ratios of their SLAs to the CS.
Once a novel configuration is decided, i or to a greater extent than of the next operations are performed every bit the organization changes to the novel configuration: (i) changing the original replica, (ii) adding or removing secondary replicas, together with (iii) changing the synchronization periods betwixt original together with secondary replicas.
+ Adjust the Synchronization Period
+ Add/Remove Secondary Replica
+ Change Primary Replica
+ Add Primary Replica
For example, for a missed subSLA amongst strong consistency, 2 potential novel configurations would be: (i) creating a novel replica nigh the customer together with making it the solo original replica, or (ii) adding a novel original replica nigh the customer together with making the organization run inward multi-primary mode. As some other example, for a social network application that spans Brazil together with India, the CS may arrive at upwardly one's heed to swap the original together with secondary replica roles to amend utility. During peak times inward India, the secondary replica inward Southern Asia becomes the original replica. Likewise, during peak times inward Brazil, the replica inward Brazil becomes primary.
The constraints for configurations may include (i) Geo-replication factor, (ii) Location, (iii) Synchronization period, together with (iv) Cost constraints. With the toll constraint, the application developers tin dot how much they are willing to pay (in damage of dollars) to switch to a novel configuration. For instance, i possible configuration is to pose secondary replicas inward all available datacenters. While the gained utility for this configuration would hold upwardly great, the toll of this configuration is probable unacceptably large.
For each configuration possibility that meets the constraints, the CS computes the expected gained utility together with the toll of reconfiguration. The CS considers the next costs for a novel configuration:
+ Storage: the toll of storing a tablet inward a detail site.
+ Read/Write Operation: the toll of performing read/write operations.
+ Synchronization: the toll of synchronizing a secondary replica amongst a original one.
Finally the CS choses the configuration that offers the highest utility-to-cost ratio, together with executes the reconfiguration operations required to transition to the novel one.
Fast Mode. When a customer is inward fast mode, read together with single-primary write operations involve a unmarried round-trip to i selected replica. No additional overhead is imposed on these operations. Multi-primary write operations utilization a three-phase protocol inward fast or wearisome mode.
Slow Mode. Slow vogue has no acquit upon on read operations amongst relaxed consistency. On the other hand, since read operations amongst strong consistency must ever choke to a original replica, the customer needs to validate that the responding replica to a strong-consistency read is yet a original replica. If not, the customer retries the read operation. Write operations are to a greater extent than involved when a customer is inward wearisome mode. Any customer inward wearisome vogue that wishes to execute a write functioning on a tablet needs to convey a non-exclusive lock on the tablet's configuration earlier issuing the write operation. On the other hand, the CS needs to convey an exclusive lock on the configuration if it decides to modify the laid of original replicas. This lock physical care for is required to ensure the linearizability of write operations.
SEA is South Eastern Asia together with WEU is West Europe.
This newspaper introduces a followup system, called Tuba, where the configuration is non fixed, together with tin hold upwardly changed on the fly. Tuba extends Pileus to address the occupation of finding an optimal configuration of original together with secondary replicas that maximizes the overall utility together with minimizes the toll for the application.
Tuba extends Pileus amongst a configuration service (CS) delivering the next capabilities:
1. performing a reconfiguration periodically for dissimilar tablets, and
2. informing clients of the electrical flow configuration for dissimilar tablets.
Data is organized into tablets every bit access units, together with each tablet is assigned to original together with secondary nodes. Of course of written report dissimilar tablets may hold upwardly assigned to dissimilar original together with secondary nodes. In guild for the CS to configure a tablet's replicas to maximize the overall utility, the CS must hold upwardly aware of the means the tablet is beingness accessed globally. Therefore, all clients inward the organization periodically post their observed latency together with the hitting together with missy ratios of their SLAs to the CS.
Once a novel configuration is decided, i or to a greater extent than of the next operations are performed every bit the organization changes to the novel configuration: (i) changing the original replica, (ii) adding or removing secondary replicas, together with (iii) changing the synchronization periods betwixt original together with secondary replicas.
Configuration service (CS)
The CS is a centralized service. To amend utility, the CS selects a novel configuration past times get-go generating all configurations that satisfy a listing of defined constraints. To permute configurations, it is costless to utilization i of the next operations:+ Adjust the Synchronization Period
+ Add/Remove Secondary Replica
+ Change Primary Replica
+ Add Primary Replica
For example, for a missed subSLA amongst strong consistency, 2 potential novel configurations would be: (i) creating a novel replica nigh the customer together with making it the solo original replica, or (ii) adding a novel original replica nigh the customer together with making the organization run inward multi-primary mode. As some other example, for a social network application that spans Brazil together with India, the CS may arrive at upwardly one's heed to swap the original together with secondary replica roles to amend utility. During peak times inward India, the secondary replica inward Southern Asia becomes the original replica. Likewise, during peak times inward Brazil, the replica inward Brazil becomes primary.
The constraints for configurations may include (i) Geo-replication factor, (ii) Location, (iii) Synchronization period, together with (iv) Cost constraints. With the toll constraint, the application developers tin dot how much they are willing to pay (in damage of dollars) to switch to a novel configuration. For instance, i possible configuration is to pose secondary replicas inward all available datacenters. While the gained utility for this configuration would hold upwardly great, the toll of this configuration is probable unacceptably large.
For each configuration possibility that meets the constraints, the CS computes the expected gained utility together with the toll of reconfiguration. The CS considers the next costs for a novel configuration:
+ Storage: the toll of storing a tablet inward a detail site.
+ Read/Write Operation: the toll of performing read/write operations.
+ Synchronization: the toll of synchronizing a secondary replica amongst a original one.
Finally the CS choses the configuration that offers the highest utility-to-cost ratio, together with executes the reconfiguration operations required to transition to the novel one.
Client execution modes
Clients withdraw to avoid 2 potential security violations: (i) performing a read functioning amongst strong consistency on a non-primary replica, or (ii) executing a write functioning on a non-primary replica. Based on the freshness of a client's view, the customer is either inward fast or wearisome mode. A customer is inward the fast vogue for a given tablet if it knows the locations of original together with secondary replicas, together with it is guaranteed that the configuration volition non modify inward the nigh future. Whenever a customer suspects that a configuration may cause got changed, it enters wearisome vogue until it refreshes its local cache.Fast Mode. When a customer is inward fast mode, read together with single-primary write operations involve a unmarried round-trip to i selected replica. No additional overhead is imposed on these operations. Multi-primary write operations utilization a three-phase protocol inward fast or wearisome mode.
Slow Mode. Slow vogue has no acquit upon on read operations amongst relaxed consistency. On the other hand, since read operations amongst strong consistency must ever choke to a original replica, the customer needs to validate that the responding replica to a strong-consistency read is yet a original replica. If not, the customer retries the read operation. Write operations are to a greater extent than involved when a customer is inward wearisome mode. Any customer inward wearisome vogue that wishes to execute a write functioning on a tablet needs to convey a non-exclusive lock on the tablet's configuration earlier issuing the write operation. On the other hand, the CS needs to convey an exclusive lock on the configuration if it decides to modify the laid of original replicas. This lock physical care for is required to ensure the linearizability of write operations.
Evaluation
They implemented Tuba to extend Microsoft Azure Storage, amongst wide consistency choices (as inward Bayou), consistency-based SLAs (as inward Pileus), together with a novel replication configuration service. Their evaluation compared amongst a organization that is statically configured. An experiment amongst clients distributed inward datacenters some the basis shows that reconfiguration every 2 hours increases the fraction of reads guaranteeing strong consistency from 33% to 54%. This confirms that automatic reconfiguration tin yield substantial benefits which are realizable inward practice.SEA is South Eastern Asia together with WEU is West Europe.
0 Response to "Paper Summary: A Self-Configurable Geo-Replicated Cloud Storage System"
Post a Comment