[SERVER-17798] Possible performance regression using SSL Created: 30/Mar/15 Updated: 30/Mar/15 Resolved: 30/Mar/15 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Security |
| Affects Version/s: | None |
| Fix Version/s: | None |
| Type: | Bug | Priority: | Major - P3 |
| Reporter: | Robert Guo (Inactive) | Assignee: | Robert Guo (Inactive) |
| Resolution: | Incomplete | Votes: | 0 |
| Labels: | 32qa | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Environment: |
Windows 2012 R2 in a hyper-V VM, 2 vCPU, 4GB ram, 128GB HDD. host machine is Intel i7 4970k, 32GB ram, 4 physical hard drives. |
||
| Attachments: |
|
| Operating System: | Windows |
| Steps To Reproduce: | hammer.mongo insert workload, but may need to run for a few hours to let the insert performance stabilize, the initial difference may be up to 40% and they gradually converge. Also it's best to set the CPU affinity to 1 or 2 cores if running on AWS, otherwise network would be the bottleneck: mkdir hammer |
| Participants: |
| Description |
|
Running an insert only workload on two identical machines shows an 18% performance hit with SSL enabled over an 11h run, as indicated by the number of bytes written and the number of inserts reported by the client. There's also a 17% increase in the number of private bytes and a small but consistent 5% increase in user time CPU for the SSL enabled version. The tests are done on the same physical machine in 2 hyper-V VMs with the client on a third Ubuntu VM, so network is not the bottleneck. All the VMs are also on different physical hard drives. detailed windows performance counters and some basic analysis are attached in the excel spreadsheet. |
| Comments |
| Comment by Robert Guo (Inactive) [ 30/Mar/15 ] |
|
client logs |