memcached info



Internal implementation:


update data in DB:
If we delete data in memcached when we update data in DB, it will cause “cache stampedes”. It is if there are many clients find out the entry isn’t in memcached at the same time, all of they will try to query from DB, then it will cause performance issue. So we need to create a lock in memcached, every client will try to check whether the lock is exist. If it is not exist, it can query and delete it after finish, other clients will wait a period time to wait cache.


data fragment and for each item:
We need to make experienment for choosing cache data!

Now for this simple application, coding something like this is simply overkill, so let’s look at a more practical example: a photo sharing site that contains lists of photographs with links to detail pages.
Many such sites display paged lists, so 100 pictures in a list might be displayed as thumbnails with basic information, ten per page. Based on all of the coding we’ve done, it might seem reasonable to go get a list of one page of data from the database, cache the resulting list and then get the second page and cache that list and so on. But there’s another way to handle that.

Start by getting the list of all 100 items that make up the list and cache that. Build the paged lists from that information, rather than going back to the database for every page. Now the same list serves someone who only wants to see 5 items per page as well as some-one who wants to see 20, which is better than having to query the
database 20 times for one person and another 5 times for the other, when all we’re doing is showing the same data paged differently.

The most important thing to remember is that this is not an all-or-nothing approach. You may have data that lends itself to stor-ing nothing except lists of pointers, and then store the individual data items separately. But you might just as easily have data where it makes more sense to store lists that contain data. This is one of those things you need to experiment with and see which makes more sense for your application.

From Using Memcached



multiget hole:
Adding new memcached server didn’t help when it’s CPU bound. What it can happens is when you add more servers is that the number of requests is not reduced, only the number of keys in each request is reduced.
Ex: If you have send 50 requests to 2 memcached servers, and each request use multi_get to get 100 entries. So each memcached server will contain 50 entries. After you add one memcached server, the requests are still 50, but only each memcached contain 33 entries. Requests are not reduced. As a result we’ve done absolutely nothing to reduce the usage of our scarce resource which is CPU!


Now I just don’t have that mood for birthday…I just don’t know what’s the birthday meaning for me now. As a result, I don’t expect a birthday party or something. But I will remember whatever you do for me, thanks forever

Perception change the solution!

今天看到一個有趣的演講, 是這樣子的, 現在想要縮短倫敦到巴黎的時間, 這時候工程師就開始想solution, 最後終於想出來了一個好方法, 這個solution花了60億, 也就是再蓋一條新的鐵路, 縮短了40分鐘的車程, 但是其實我們可以往另一個方向去想,  我們用30億的錢把全世界最頂尖的model都請來站在鐵軌上, 這樣我相信大家還會要求把列車速度降下來呢! The same problem, but different solution!

還有一個故事, 有一個國王想要他的子民能夠推廣多種馬鈴薯, 但是那個時候沒有人想種這種看起來髒髒的作物, 但是他又不能強迫大家種, 也不能把不種的人處決掉, 所以他就散佈假消息出去, 說馬鈴薯是貴族愛用的食物, 然後再把如何栽種的方法寫下來, 再想辦法讓別人把方法偷走~ 果不其然, 後來大家就開始種植馬鈴薯!

其實很多時候很多東西都只是認知上的差異, 這也是為什麼安慰劑(placebo)的存在, 如果我們能給予舊的東西新的象徵, 也會有不一樣的結果~

如果不告訴你喝的酒很貴, 你真的分的出來嗎??

曾經再書上看過一個例子, 你覺得一瓶水應該賣多少錢呢??

在問一下, 你覺得一瓶水對於一個在沙漠的人, 又應該賣多少錢呢?

怎麼樣充分利用你所有的東西, 這可真是一門學問阿~