Hi Martyn, The Java interface to the Memcache API, unfortunately, is noticeably slower than the Python version. This is the cause of the discrepancies you're seeing here. As always, using the batch interface to minimize round trips is a good idea.
Please do file a bug, though - the discrepancy between runtimes is something we'd like to address. -Nick Johnson On Fri, Aug 7, 2009 at 8:58 AM, Martyn<[email protected]> wrote: > > After some more detailed logging I have isolated the problem and it is > worrying. > > It so happens that I make several memcache requests, I had assumed > that since I was making > calls to the Datastore API, any delay would be due to that, but it > turns out that around > half the delay is due to the memcache calls. > > Specifically, I am seeing ~13ms latency for each memcache (single get/ > put/increment), and since I am > making typically 5 calls, I am seeing ~65ms delay due to this. > > This is in contradiction with AppEngine System status that indicates > memcache latency ~4-5ms. > > Of course, what is meant by latency vs call elapsed is a little > ambiguous, but the figures for Datastore latency > tie up with what I see in my request logs. > > In other words I time the api call/return and I see around the same > figures for Datastore latency but almost three times the memcache > latency. > > Is this a problem specific to the Java API perhaps? > > If I batch three memcache puts using putAll, I see 13ms for the three > rather than 39ms. > > On 30 July, 19:47, Martyn <[email protected]> wrote: >> I am using the low-level Java Datastore interface and writing a >> "transactional" store - where each entity stores a number of data >> updates in a single Blob. >> Each transaction requires two updates, one to a Header entity, and one >> to the new transaction data entity. Keys for each are created >> explicitly. This has a number of advantages, specifically with regard >> to guaranteed transaction isolation and caching potential. >> However, I am disappointed with the insertion performance. I use the >> DatastoreService put(Iterable) to invoke the two updates witha single >> API call (or rather one update and one insert), but I am seeing >120ms >> request times in the logs. The log trace confirms the delay is largely >> down to the datastore put. >> >> Am I missing something, or is this really the best I can hope for? >> Apart from the blob data, which is normally ~1k, a couple of debug >> properties (Date and VM hash) are included but set using >> setUnindexedProperty. >> >> Any help/advice appreciated. > > > -- Nick Johnson, Developer Programs Engineer, App Engine --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Google App Engine" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en -~----------~----~----~----~------~----~------~--~---
