[05:06] Hello === Guest81 is now known as Tom [05:07] 有人吗 === Tom is now known as Guest25885 [05:10] hi === Guest81 is now known as yekjhhee === Guest81 is now known as yekjhhee [05:18] hi [05:19] hi [05:20] 有人吗 [05:21] 有 [05:21] 居然真的有人 [05:21] 感觉发现了新天地 [05:58] 感觉大清要亡了啊 [06:19] contrun[m]: cherrot: zao. [06:20] imadper 大清什么时候亡啊 [06:20] contrun[m]: tomorrow. [06:21] imadper Great 磨难要结束了 [06:21] contrun[m]: How? [06:21] imadper 活着没劲啊 [06:22] contrun[m]: Try to find a condom-friend? [06:22] imadper 刁夶一日不完我一日不找 condom-friend [06:23] contrun[m]: Why 刁夶 can't be your condom friend? [06:23] * imadper Shitty sogou-pinyin [06:23] imadper 刁明泽可以 [06:23] imadper rime 大法好 [06:24] contrun[m]: Is it possible that hijack read/write with ptrace to implement a cache system for NFS? [06:24] contrun[m]: I mean, a local arc for nfs. [06:25] contrun[m]: How to handle mmap then? [06:25] imadper 你他妈这是要在错误的道路上越走越远 [06:25] contrun[m]: More thoughts? [06:26] contrun[m]: more details? [06:26] imadper 不是说了用 ceph 吗 [06:27] contrun[m]: shitty ceph. I don't even know if ceph has a local cache [06:28] imadper There is no harm in trying [06:28] contrun[m]: Hard to measure. [06:28] contrun[m]: cache is always transparent [06:29] imadper 你想要达到什么效果 [06:29] 明明是多难兴邦 [06:29] contrun[m]: with local zfs, arc hits 99% on our workload. [06:29] contrun[m]: With 20G arc. [06:30] 2.10封路能结束吗? [06:30] contrun[m]: Which provides 600K iops which 16K recordsize. [06:30] albet70 你可能还活在梦里面 [06:30] contrun 中国梦 [06:31] imadper 具体使用场景的效果 [06:32] 等刁夶亲自翠吧 [06:32] contrun[m]: When runing filecoin action0, there is nearly zero physical disk io with arc. Arcstats shows that arc covers almost all of the read. [06:33] imadper 访问的文件是网络或本地缓存在内存？ [06:34] contrun[m]: https://sm.ms/image/DosbV4drQaPtyqw currently, in local. However I need to move it to lan. [06:35] imadper 缓存的内存不是 kernel 的 buffer 你应该搜 fs-agnostic buffer [06:36] contrun[m]: I just need ceph or nfs have similar cache. [06:36] contrun[m]: And if ceph has built-in memory cache, I'd use it directly. [06:37] 都啥时候了，还整天来些心灵鸡汤的新闻， [06:37] 还有那些没事蹭热点的院士 [06:37] imadper ceph 和 nfs 的一般缓存是指的缓存到本地硬盘啊 [06:38] albet70 大清要亡了 [06:38] 这次被人实名举报，干脆这个锅就让武汉病毒研究所和中科院那些院士背了吧，不杀不足以平民恨 [06:38] contrun[m]: that's why I need to hijack read/write with ptrace. [06:39] contrun[m]: To reduce hardware cost, we have no nvme on ever worker. [06:39] albet70 这个显然是无稽之谈 [06:39] s/er/ery/ [06:40] imadper linux 的 buffer 是怎么触发的？“ [06:40] 管它是真是假，总有人需要背锅吧 [06:40] contrun[m]: readahead? [06:41] imadper 调 linux buffer 才是正确的方向 [06:41] 平时拿皇粮的，此时就是他们为国捐躯的时候 [06:41] albet70 让亲自指挥亲自部署的刁夶上 [06:42] contrun[m]: is linux buffer a LRU? [06:42] albet70 来一个罪己诏那是坠吼的 [06:42] 网上都说这次问题这么大，就是因为年前那个女的开车碾压了龙脉 [06:43] 我感觉这个说法非常科学 [06:43] 应该把这个女的全族祭祀了 [06:43] imadper 母鸡啊 [06:44] albet70 大帝危矣 [06:45] 你想想龙脉那是什么地方，竟然被这种妖精开车碾压，这个国运都被搞没了 [06:45] albet70: Agree with you. [06:47] albet70 你国国运不早就没了刁夶都能上 [06:51] imadper 你可以看下数据库是怎么优化的数据库和你的使用场景很像 [06:59] imadper 不行数据库可能没有用比较通用的缓存方案 rocksdb 好像是有 writeahead log 数据本身就存在内存到了一定大小就 flush 到硬盘 [07:00] contrun[m]: yep, that's how lsm tree works. [07:01] contrun[m]: However, I need it for read. [07:01] 北京下雪了？ [07:02] 天津也下雪了？ [07:02] contrun[m]: https://docs.ceph.com/docs/giant/rbd/rbd-config-ref/ [07:02] albet70 下了雪地里一个小画家都没有 [07:02] imadper 不错 [07:05] imadper 数据库读取硬盘文件也是有缓存的不知道是怎么做的 [07:06] contrun[m]: LRU or MFU [07:06] contrun[m]: arc is the combin of LRU and MFU [07:07] contrun[m]: it adjust the size of LRU and MFU automaticlly to fix kinds of workload. [07:09] contrun[m]: https://sm.ms/image/dEsAzRTiabnBMlg in my case, MFU works well. [07:16] imadper 你需要的是一个 tmpfs 作为前端 nfs or whatever 作为后端的 https://bcachefs.org/ 还有两个问题 bcachefs 有没有 failure recovery tmpfs 要是满了 bcachefs 会怎么办 [07:17] contrun[m]: IIRC bcachefs works at block layer, [07:17] https://en.wikipedia.org/wiki/CacheFS [07:17] contrun[m]: cachefs uses another fs as cache. not memory. [07:18] tmpfs [07:19] contrun[m]: https://sm.ms/image/ZNzGfU8Ee7OKstu [07:20] contrun[m]: https://www.redhat.com/archives/linux-cachefs/2011-October/msg00005.html [07:20] contrun[m]: :-) [07:20] imadper 总有一款 cachefs 适合你 [07:20] contrun[m]: "You cannot use tmpfs because it doesn't support xattr needed for fscahe." [07:22] contrun[m]: I would give ceph a try. [07:23] contrun[m]: However all hdds are managed by zfs now. I can't destory zpool now. [07:23] imadper 是啊你他妈为什么要瞎几把折腾呢 [07:23] contrun[m]: just know that ceph has a local memory cache. [07:24] contrun[m]: and still not sure if it's a mfu. [07:24] imadper premature optimization [07:24] imad [07:24] imadper 又不是不能用 [07:25] contrun[m]: Just don't want to take too much time on testing ceph/tmpfs/bcachefs [07:25] imadper 你的问题加内存可以解决？ [07:25] contrun[m]: Nope, The program is alread runing. The job I'm doing now is find out the fastest storage solution. [07:25] contrun[m]: which means, I'm in optimization phase now. [07:26] contrun[m]: It's not "premature" [07:26] imadper 牛逼 [07:26] contrun[m]: Increasing hardware cost is possible iff I have a sane caching system to use the memory. [07:27] contrun[m]: So I'm finding the `sane caching system` [07:28] imadper 改个缓存策略会这么难？ [07:28] contrun[m]: how? [07:29] contrun[m]: Show me the command. [07:29] imadper 没有我跳楼你举证我跳楼 [07:29] contrun[m]: you jump, I jump [07:31] imadper 别你跳了你老婆怎么办你们这些人啊怎么一点都不顾家呢 [07:31] contrun[m]: jump jump more health [07:32] imadper jump jump healthier [07:32] contrun[m]: .... Okay [08:01] kubuntu用tightvncserver，键盘输入乱码是咋回事儿啊 [08:01] 完全是乱的😂 [08:12] xrdp就更不行，连接上就显示个鼠标箭头…… [08:22] 有么有然滴嘛？ [08:32] ... === sixand1 is now known as sixand === sixand1 is now known as sixand === sixand1 is now known as sixand [16:03] 好困啊