Take the 2-minute tour ×
Database Administrators Stack Exchange is a question and answer site for database professionals who wish to improve their database skills and learn from others in the community. It's 100% free, no registration required.

I've searched the web for a little while and haven't found an answer I'm confortable with. I am testing a SQL Server on a VM that was presented new disc.

If disk latency is reduced, shouldn't throughput increase? Is there a direct correlation between disk latency and throughtput? Seems like if I have faster disk I should see an increase in throughput.

share|improve this question
    
In theory there is no difference between theory and practice. In practice there is. –  swasheck Mar 7 at 19:15
    
Latency on disk or latency between OS and storage? –  swasheck Mar 7 at 19:16
1  
to answer this properly we really need a bit more detail about how you are using the terms 'latency', 'throughput', 'faster' and even 'disk'. For example is the disk presented to the VM a real physical disk or a backing file on the host (or something else). –  Jack Douglas Mar 7 at 19:20
    
It is physical disk. Latency I am reffering to Avg. Disk sec/Read & Avg. Disk sec/Write. Throughput - number of transactions my benchmarking tool can compute. –  sqllayman Mar 7 at 20:25
add comment

1 Answer 1

up vote 7 down vote accepted

It might help to define terms up-front.

The term latency is used in (at least) two ways regarding I/O. It can mean:

  • The average time it takes to service an I/O at any given time. That's how Paul Randal is using it in "How to examine IO subsystem latencies from within SQL Server" and it can be measured using (for example) Performance Monitor with counters such as "Avg. Disk sec/Read". I'm going to call this the 'measured' latency, and it is related (inversely) to the queue length at a given time, but is a much more helpful measure because it is easy to relate to actual performance from a user perspective.
  • The average time a single I/O takes in isolation. That's how it is typically used on spec sheets for HDDs and SSDs like this one from Seagate. I'll call this the 'inherent' latency (it's also sometimes called 'access time').

Throughput can also mean different things in different contexts:

  • The maximum rate of sequential data transfer, typically measured in MB/sec
  • The maximum rate of I/O operations typically measured in IOPS
  • The current rate of data transfer such as that measured using Performance Monitor with counters such as "Disk Bytes/sec"

Seems like if I have faster disk I should see an increase in throughput.

Because of the way you phrase this, I think you are specifically talking about the quoted MB/sec and 'inherent' latency for disk drives. If that's the case then the answer is "no, there is no direct correlation between disk latency and throughput". Latency for a HDD is affected by the seek time and the spin speed, but it only applies to the first read: subsequent sequential reads do not have the same latency. The mechanics are different for SSDs, but they can still be tuned by the manufacturer for high 'throughput' or high IOPS with issues like block size being important.

share|improve this answer
    
It is physical disk. Latency as in Avg. Disk sec/Read & Avg. Disk sec/Write. Throughput as in number of transactions my benchmarking tool can compute. –  sqllayman Mar 7 at 21:11
    
by "number of transactions" you mean the same thing as IOPS? –  Jack Douglas Mar 7 at 21:11
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.