Why are the delay models different between bbssd and zns in FEMU?

[wnsah814]

Hello FEMU developers,

While examining the FEMU codebase, I noticed that the delay modeling approaches for bbssd and zns are quite different. I'm curious about the reasoning behind these differences.

Why does bbssd track delays at the LUN level while zns uses plane-level granularity?

bbssd delay model:

uint64_t ssd_advance_status(struct ssd *ssd, struct ppa *ppa, struct nand_cmd *ncmd)
{
    // ...
    struct nand_lun *lun = get_lun(ssd, ppa);
    // ...
    switch (c) {
    case NAND_READ:
        nand_stime = (lun->next_lun_avail_time < cmd_stime) ? cmd_stime : \
                     lun->next_lun_avail_time;
        lun->next_lun_avail_time = nand_stime + spp->pg_rd_lat;
        lat = lun->next_lun_avail_time - cmd_stime;
        break;
    // ...
}

zns delay model:

uint64_t zns_advance_status(struct zns_ssd *zns, struct ppa *ppa, struct nand_cmd *ncmd)
{
    // ...
    struct zns_plane *pl = get_plane(zns, ppa);
    int nand_type = get_blk(zns,ppa)->nand_type;
    
    uint64_t read_delay = zns->timing.pg_rd_lat[nand_type];
    // ...
    switch (c) {
    case NAND_READ:
        nand_stime = (pl->next_plane_avail_time < req_stime) ? req_stime : \
                     pl->next_plane_avail_time;
        pl->next_plane_avail_time = nand_stime + read_delay;
        lat = pl->next_plane_avail_time - req_stime;
        break;
    // ...
}

I'm simply trying to understand the design rationale behind these differences.
Thank you for your help!