Hi Yocto experts, I'm a Yocto newbiee and I have a concern regarding the slowness of Yocto for our team(around 100 people). There are two main reasons for this:
1. Yocto itself is slower compared to other build systems. 2. Each user in our team has their own private SSTATE_DIR and DL_DIR, which means the artifacts cannot be shared. This leads to unnecessary disk and CPU usage. 3. Every user works on SATA devices, so the IOPS and throughput are both low. Since it's hard for us to invest in expensive enterprise NVMe for all build servers, I have an idea to address this issue. My suggestion is to use Lustre/NFS/Glustrefs along with small NVMe to share everyone's SSTATE/DL_DIR. I prefer Lustre over NFS because Lustre offers superior IO performance scalability, making it suitable for data-intensive tasks. It splits the data, allowing it to be requested from one server but sent from one or more other servers. For the storage of SSTATE/DL_DIR only, the size requirement would not be high, so a small NVMe should be sufficient. NVMe can provide much higher throughput and IOPS compared to our current slow SATA, so I guess it can meet our needs. If necessary, we can easily add another NVMe to another server since Lustre is a distributed file system. The overall cost would not be significant. My questions are: 1. Are there any errors in my understanding? 2. Is this solution feasible? If not, why? If yes, can it be further improved? I look forward to hearing from you! Regards, Li
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#60962): https://lists.yoctoproject.org/g/yocto/message/60962 Mute This Topic: https://lists.yoctoproject.org/mt/101284965/21656 Group Owner: [email protected] Unsubscribe: https://lists.yoctoproject.org/g/yocto/unsub [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
