Experimental Configuration Every experiment was performed on a cluster with 3

Experimental configuration every experiment was

This preview shows page 147 - 150 out of 211 pages.

Experimental Configuration Every experiment was performed on a cluster with 3 servers. Each machine in the cluster runs a 8-core Intel Xeon E5620 processor clocked at 3.8 GHz and has 16 GB of RAM. Each server had one instance of CockroachDB node running on it and the data was replicated three ways. The experiments were run with 100,000 key-value data items without data partition, hence the experiments did not have distributed transactions. Each data point in the benchmarking process corresponds to an experiment performed with 100,000 transactions accessing 100,000 data items. 10000 warm-up reads are per- formed before starting each experimental run. Each data point reported in the results is an average of 3 repetitions of such an experiment. The benchmark is available on Github [137]. 131
Image of page 147
Dynamic Timestamp Allocation for Tackling High Contention Workloads Chapter 5 Experimental Methodology As mentioned in 5.3.2, for dynamic timestamp ordering, the creation of a transaction record happens on the first operation, be it a read or a write, rather than on the first write operation, as is the case with fixed timestamp ordering. In fixed timestamp ordering, if the transaction consisted of all read-only operations, no transaction record is created. Furthermore, the creation of the transaction record also encompasses the replication of the transaction record (to the replicas holding the range associated to the first key accessed by the transaction) using Raft, both at the begin and when the commit decision is made. Hence using this approach in a read-dominant system, adds significant delays compared to the original implementation of CockroachDB. There are many approaches to avoid the bookkeeping done by MaaT for read-only transactions. One such solution was proposed by Agrawal et al. [138], which can be implemented in CockroachDB along with MaaT. A read-frontier based on this approach proposed by Agrawal et al. [138] can be continuously maintained, which provides access to the latest commit timestamp below which no transaction can commit. The timestamp corresponding to the read frontier can be used as the timestamp to read the items accessed in the read-only transactions. In order to apply optimized solutions for read-only transactions, we first need to iden- tify if the transaction will be read-only. After a discussion with CockroachDB developers, we learnt that there is no provision as of now in the database to specify a transaction as read-only. Implementing this change will require modification in multiple layers of the database and is a complex task, and it is not in the scope of this paper. In order to overcome this issue, in our experiments, we assume that every read-only transaction is successful and will not have the overhead of creating and maintaining a transaction record. If the optimization was to be implemented, read-only transactions would hit only one server, which would respond to the client without creating and replicating transac- 132
Image of page 148
Dynamic Timestamp Allocation for Tackling High Contention Workloads Chapter 5 (a) Aborts
Image of page 149
Image of page 150

You've reached the end of your free preview.

Want to read all 211 pages?

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Stuck? We have tutors online 24/7 who can help you get unstuck.
A+ icon
Ask Expert Tutors You can ask You can ask You can ask (will expire )
Answers in as fast as 15 minutes