Throughput of read operations
jeremiah.peschka at gmail.com
Fri Aug 12 14:24:01 EDT 2011
If the numbers you reported are the average, that's interesting. Were you using an EBS volume or an ephemeral volume?
You can definitely be seeing a noisy neighbor effect, although typically you'll see inconsistent results.
Jeremiah Peschka - Founder, Brent Ozar PLF, LLC
Microsoft SQL Server MVP
On Aug 12, 2011, at 8:37 AM, Maria Neise wrote:
> I ran it on one day (on a sunday) but each workload 4 or 5 times and
> my hard drive had 150GB.
> So it could be Amazons fault? :)
> On 12 August 2011 16:50, Jeremiah Peschka <jeremiah.peschka at gmail.com> wrote:
>> One question that pops into my mind is are these numbers averages over 3 or 5 runs or is this a single run?
>> In my experience with EC2, I've run into I/O inconsistencies when I'm using anything less than 1TB drives and a large instance. There's a lot of potential for a noisy neighbor to steal I/O or CPU, or memory on the host OS which will put pressure on all other guests.
>> Jeremiah Peschka - Founder, Brent Ozar PLF, LLC
>> Microsoft SQL Server MVP
>> On Aug 12, 2011, at 2:23 AM, Maria Neise wrote:
>>> I am doing some benchmarks with Riak and set up a cluster with 6
>>> server using Amazon EC2. On each server I have GB of data. I am using
>>> Bitcask as backend and the Java-API for my client. I have some
>>> workloads with different proportion of operations, for example insert,
>>> read and update. I started a workload with 95% read and 5% update and
>>> achieved a throughput of 1207 operations/second. When I tried out
>>> another workload with 100% read I only got 228 ops/sec. But I thought
>>> Riak would be faster with reads than with updates. I would appreciate
>>> if you could give me a hint, why Riak achieves a higher throughput
>>> when doing updates and inserts mixed with reads.
>>> Thank you in advance.
>>> riak-users mailing list
>>> riak-users at lists.basho.com
More information about the riak-users