Performance Regression after increase in high/low watermarks

Dipti Borkar added a comment - 17/Jan/13 1:38 PM Ronnie, please attached your reports from performance runs. Assigning this to Jin right now because he's been driving the capacity planning stuff of late, please re-assign as required.

Jin Lim added a comment - 21/Jan/13 1:23 PM Ronnie, when data is available please attach your reports from the performance runs here. Also, for better understanding, let's be specific in what rate of performance degradation we are talking about here. Please assign it back to me after you have attached the requested reports.

Ronnie Sun added a comment - 21/Jan/13 1:39 PM With higher water mark settings: - low cache miss (1%) looks good/ - higher cache miss tests with high mem_quota( ~ 93%) is worse, which might be causing by high swap usage. Attached some graphs. QA team will test (especially with 2.0.1 build) and confirm.

Farshid Ghods added a comment - 22/Jan/13 2:23 PM per bug scrub - assigning back to Jin

Jin Lim added a comment - 24/Jan/13 1:04 PM From the latest round of performance test (Ronnie), high/low water mark changes seem to be OK. However, we would like to still wait for high cache miss ratio test and confirmation from QE (Ketaki).  

Jin Lim added a comment - 24/Jan/13 1:06 PM Ketaki, please assign it back to Jin or mark it as resolved once your test (high cache miss) completed with concrete result. Thanks!

Ketaki Gangal added a comment - 28/Jan/13 1:18 PM Not seeing any system-regression on build 2.0.1-140 so far. No higher swap w/ higher cache-miss ( 5-20 %)

Farshid Ghods added a comment - 28/Jan/13 3:43 PM Ketaki can you please post the graph which shows that there is not much swap usage int he cluster after X days also it helps if you print out the tcmalloc fragmentation output here priori to resolving the issue

Ketaki Gangal added a comment - 28/Jan/13 4:05 PM Refer page-10 for swap stats. Swap on the current cluster did not grow w/ the change in High/Low Watermarks.