← All insights
RESEARCHZcash Forum · jatinsahijwani

Shielded wallet performance on low-end Android phones

Before measuring wallet performance on phones popular in the Indian market, Zcash India pre-registered its methodology, promising to publish all results, including unfavorable ones.

Zcash India published methodology 1.0 for a benchmark measuring shielded wallet performance on low-end Android devices before any data collection. It started from the problem that no public data exists on shielded wallet performance on the devices most used in the Indian market. It committed in advance to publish all results under CC0, including unfavorable ones.

Measurement takes only two runs per device. One is a cold sync from seed restore to full sync, and the other is one shielded send. During these two runs it automatically records sync time, send time, proving time, peak memory, OOM kills, heat, battery, mobile data usage and failure types. All figures are published as single measurements, without averaging.

The targets are 20 Android devices in the ₹6,000 to ₹12,000 range, covering 2 to 6GB of RAM and Android 11 to 15. The primary wallet is Zodl, and on 6 devices spanning the RAM range one other wallet is compared under the same protocol. The reason is that Zcash India contributes to Zodl, so a single-wallet study cannot serve as a benchmark. All runs use the wallet's default light client server.

Besides the controlled runs, field runs are also done at events on the devices of consenting attendees. In the field it records sync progress at the 10-minute mark, and the two kinds of results are reported separately and not combined. Attendees' own wallets are not used, and no personal information such as addresses, seeds or balances is recorded. A sync over 120 minutes is recorded as sync_timeout.

The three raw-data CSVs (results, excluded, devices) and the results report will be published on the forum by November 30, 2026. Excluded runs are also published with their reasons, and no leaderboard or composite score is created. It also stated its own limitations: the 20 devices are a matrix, not a representative sample, and with single runs small differences are meaningless. It also disclosed conflicts of interest: a Zodl translation PR and an event supported by ZCG.

This summary is based on the original; check the original for exact details.