RFR(S): 8020151: PSR:PERF Large performance regressions when code cache is filled

[Igor Veresov]

It's worth a try. But the "hotness" logic is probabilistic and imprecise - it will notice only methods that are on stack during safepoints. Those are going to be pretty narrow snapshots of activity. I suspect that for large flat profiles (like enterprise apps and friends) you could be missing methods that are rather warm in reality, which will cause recompilation oscillations. Although if the statistics are allowed to accumulate enough may be it's going to work out, an experiment will tell. The patch is a good start, the stack sampling, IMO, is totally the right approach for filtering out the hot methods.

The "disconnect" logic on the other hand is sort of precise. Although, looking now at the code it's not quite clear to me how it works, it doesn't seem to be any patching going on to divert the control for the cases when the nmethod is called directly or through an IC. So I guess it's not really a full disconnect?  Anyways, in theory, with some work, we can make the disconnect logic to precisely measure the time the method is inactive. Which should provide precise information about the warm/cold methods.

Btw, also just noticed a bunch of flaws in the interaction of the disconnect logic and tiered. The nmethod's "reconnection" happens in CompileBroker::compile_method(), which firstly will be called by tiered only after a rather substantial number of invocations in the interpreter (up to 128), and secondly will be subject to all the prioritization rules (it probably should not), and also we don't check if the comp level of the reanimated nmethod matches the request. If the disconnect logic is to stay, the interpreter should be able to know if the method has saved code and be able to call into the runtime immediately to reanimate it.

igor

 
On Aug 21, 2013, at 10:27 PM, Albert Noll <albert.noll at oracle.com> wrote:

> Hi Igor,
> 
> thanks for looking at the patch. Actually, I think - just as Vladimir pointed out - that we can get rid of
> the "disconnect" logic. We now have the hotness of a method and if the code cache fills up, we and
> we decide to schedule the method for removal, we set it to not_entrant. 
> It seems that adding the method to the list of disconnected methods just buys a little more time until we decide to make the method not-entrant. However, we can have the same effect by setting the  threshold differently.
> 
> What do you think?
> 
> Best,
> Albert
> 
> On 22.08.2013 10:02, Igor Veresov wrote:
>> May be instead of "(_traversals > _last_flush_traversal_id + 2)" we should timestamp a method when it's disconnected, and then use a rule like if a method has been disconnected for k * reverse_free_ratio() seconds then it's ok to kill it. We can also sort the nmethods that pass that filter by the amount of time they were disconnected and select most likely candidates for flushing. This should allow to basically do disconnect/flush in every traversal, which should make things faster. Timestamps would be obtained only once per traversal or something like that. What do you think?
>> 
>> Pretty cool idea to reverse-prioritize disconnects on hotness.
>> 
>> igor
>> 
>> On Aug 21, 2013, at 4:42 AM, Albert Noll <albert.noll at oracle.com> wrote:
>> 
>>> Hi all,
>>> 
>>> could I have reviews for this patch? Please note
>>> that I do not yet feel very confident with the sweeper,
>>> so please take a close look.
>>> 
>>> jbs: https://jbs.oracle.com/bugs/browse/JDK-8020151
>>> webrev: http://cr.openjdk.java.net/~anoll/8020151/webrev.00/
>>> 
>>> 
>>> Many thanks in advance,
>>> Albert
>>> 
>>> 
>>> Problem: There can be large performance regressions when the code cache fills up. There are 
>>> several reasons for the performance regression: First (1), when the code cache is full and methods 
>>> are speculatively disconnected, the oldest methods (based on compilation ID) are scheduled for
>>> flushing. This can result in flushing hot methods. Second (2), when compilation is disabled due to a full
>>> code cache, the number of sweeps can go down. A lower number of sweep operations results 
>>> in slower method flushing.
>>> 
>>> Solution:
>>> Introduce a hotness counter that is set to a particular value (e.g., 100) when there is an activation
>>> of the method during stack scanning. The counter is decremented by 1 every time the sweeper 
>>> is invoked.
>>> 
>>> ad (1):
>>>   A VM operation that speculatively disconnects nmethods, selects the methods that should be
>>>   flushed based on the hotness. For example, if 50% of the code cache shall be flushed, we flush 
>>>   those methods that have not been active while stack scanning for the longest time. Note that 
>>>   while this strategy is more likely to flush cold methods, it is not clear to what extent the new 
>>>   strategy fragments the code cache.
>>> 
>>>   Changes in NMethodSweeper::speculative_disconnect_nmethods(bool is_full)
>>> 
>>> ad (2)
>>>   Currently, methods are removed from the code cache if:
>>>     a) code cache is full 
>>>     b) class is unloaded 
>>>     c) method is replaced by another version (i.e., compiled with a different tier) 
>>>     d) deopt
>>> 
>>>    The current patch adds a 5-th possibility to remove a method from the code cache. 
>>>    In particular, if a method has not been active during stack scanning for a long-enough 
>>>    amount of time, the method is removed from the code cache. The amount of time
>>>    required to flush the method depends on the available space in the code cache. 
>>>    
>>>    Here is one example: If a method was seen on a stack the hotness counter 
>>>    is set to 100. A sweep operation takes roughly place every 100ms. I.e., it takes 
>>>    100ms * 100 = 10s until the hotness counter reaches 0. The threshold that determines
>>>    if a method should be removed from the code cache is calculated as follows:
>>>  
>>>    threshold = -100 + (CodeCache::reverse_free_ratio() * NMethodSweepActivity)
>>> 
>>>     For example, if 25% of the code cache is free, reverse_free_ratio returns 4.
>>>     The default value of NMethodSweepActivity is 10. As a result, threshold = -60.
>>>     Consequently, all methods that have a hotness value smaller than -60 (which 
>>>     means they have not been seen on the stack for 16s) are scheduled to be flushed
>>>     from the code cache. See an illustration of the threshold as a function of the available
>>>     code cache in threshold.pdf
>>> 
>>>     Note that NMethodSweepActivity is a parameter that can be specified via a -XX
>>>     flag.
>>> 
>>> Changes in NMethodSweeper::sweep_code_cache()
>>> 
>>> 
>>> A very preliminary performance evaluation looks promising. I used the DaCapo 
>>> benchmarks where a series of benchmarks is executed in the same VM instance.
>>> See performance.pdf . The x-axis shows the benchmarks. Assume we have 2 benchmarks 
>>> (BM). The execution sequence is as follows:
>>> 
>>> BM1 (Run 1-1)
>>> BM1 (Run 2-1)
>>> BM2 (Run 1-1)
>>> BM2 (Run 2-1)
>>> 
>>> BM1 (Run 1-2)
>>> BM1 (Run 2-2)
>>> BM2 (Run 1-2)
>>> BM2 (Run 2-2)
>>> 
>>> 
>>> A value larger than 0 on the x-axis indicates that the version including the proposed patch is faster.
>>> I.e., the values are calculated as follows: (T_original / T_with_patch) - 1. T is the execution time 
>>> (wall clock time) of the benchmark. ReservedCodeCacheSize is set to 50m.  I used three runs and 
>>> the arithmetic average to compare the numbers. I know, we need much more data, however, 
>>> I think we can see a trend.
>>> 
>>> The current patch does not trigger a warning that the code cache is full and compilation has been
>>> disabled.
>>> 
>>> Please let me know that you think.
>>> <threshold.pdf><performance.pdf>
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20130822/cb3d1dd7/attachment.html